economics the determinants and effectiveness of industrial …€¦ · abstract: industrial policy...
TRANSCRIPT
ECONOMICS
THE DETERMINANTS AND EFFECTIVENESS OF INDUSTRIAL POLICY IN CHINA: A STUDY BASED
ON FIVE-YEAR PLANS
by
Yiyun Wu Zhejiang University, China
Xiwei Zhu
Zhejiang University, China
and
Nicolaas Groenewold Business School
University of Western Australia
DISCUSSION PAPER 16.21
THE DETERMINANTS AND EFFECTIVENESS OF
INDUSTRIAL POLICY IN CHINA: A STUDY BASED ON
FIVE-YEAR PLANS
Yiyun Wu
(Zhejiang University, China)
Xiwei Zhu
(Zhejiang University, China)
Nicolaas Groenewold
(The University of Western Australia, Australia)
Version: 201608
DISCUSSION PAPER 16.21
Abstract: Industrial policy is ubiquitous in both developing and developed countries.
In China, government intervention in favor of specific industries is believed to play an
important role in economic development. The targeting of strategic industries for
preferential treatment has been a central part of both national and provincial
Five-Year Plans (FYPs). While arguments for and against industrial policy are
well-known, empirical evidence regarding both the formulation and effects of such
policy are scant. Our paper contributes to the literature by examining industrial policy
in China over the period of four FYPs (the 9th to the 12th, running over the period
1996 to 2015). In particular, based on China’s national and provincial FYPs, this
paper investigates the determinants and effectiveness of industrial policy at the
four-digit manufacturing sector level. This paper finds that the central government’s
preferences act as a key determinant of the provincial governments’ FYPs. High
productivity does not improve an industry’s chance of selection by the provincial
governments. We also show that preferential policy has a significant positive effect on
industry output growth while the FYP is in effect but that there is no evidence of a
beneficial effect beyond the end of the particular FYP.
Keywords: Industrial Policy; Five-Year Plan; Industrial Development; Productivity
1
1. Introduction Industrial policy (IP) is widespread in both developing and developed economies. Although
economists have held different opinions about the role of IP, policy-makers of many developing
countries believe that the developing world is full of market failures and it is natural for their
governments to seek to escape the poverty trap through forceful industrial interventions. Thus
industrial policy, which comprises a variety of actions designed to correct market failure and
foster the reallocation of productive factors from low to high productivity activities, is widely
observed during the process of structural change (Pack, 2000, Pack and Saggi, 2006; Rodrik, 2004,
2009). This is also true for China where it has been an important part of Five Year Plans.
Although China has implemented market-based economic reforms and transitioned from a
Soviet-style planned economy since 1978, the far-reaching national programs officially titled
“Five Year Plan for National Social and Economic Development” (henceforth FYP) still provide
targets and programs for the country’s next five years of development and are believed to continue
to have a substantial impact on the national economic landscape.
Besides the national FYP (NFYP), each provincial government in China also issues its own
five-year program for regional development. In general, the provincial FYP (PFYP) imitates social
and economic targets of the NFYP plus other development targets determined by local economic
conditions. Special treatment (e.g. tax credits, subsidies, and favorable loan rates) by government
at all levels is then assigned to strategic sectors identified by the national and provincial FYPs. For
example, in the 12th NFYP (2011-2015), there were three main priorities viz., sustainable growth,
industry upgrading and promotion of domestic consumption. These three priorities were copied by
most PFYPs and induced provincial governments to reallocate productive resources to
manufacturing sectors such as energy, machinery, automotive, IT, infrastructure, and
biotechnology, which had already been selected as pillar industries in the 12th NFYP. In this way,
the NFYP is implemented by the actions of provincial governments by which they not only show
their loyalty but also obtain extra development resources from the central government by
following the NFYP. Therefore the FYP is one of the most significant indicators of whether and
how regional governments take part in resource reallocation in an attempt to influence the
economic performance of industries in the short and long term. An analysis of this relationship
between NFYPs and PFYPS which is the first main concern of this paper.
A central part of IP in many economies (including China) is to pick particular industries for
preferential treatment. This policy begs the question: does it work; that is, do the selected
industries benefit from the policy and, more broadly, does such policy benefit the economy as a
whole? In general, many economists are skeptical of the efficacy of IP. They suspect governments
are not very good at “picking winners” and that the real aim of IP is to benefit friends, electors in
marginal electorates, lobby groups and so on. Nevertheless, there are some good arguments based
on the presence of market failure to support the use of IP; these are comprehensively reviewed by
Rodrik (2004) and Pack and Saggi (2006).
2
One argument for IP is based on the presence of market imperfections. Dethier et al. (2011),
for example, pointed out that firms in emerging markets are exposed to more severe constraints
due to financial market frictions than those in the developed countries, and financial constraints
tend to be one of the primary obstacles for firms to invest in developing countries. Policies such as
subsidies for innovation are capable of providing low-cost capital to relax firms’ financial
constraints in developing countries.
A second rationale for IP is the infant industry argument, which was first formulated by
Hamilton (1791) and List (1856). The justification for infant-industry protection is the presence of
dynamic learning effects that are external to the firm. The Mill-Bastable Test has been formulated
to justify this kind of protection policy (Mill, 1848; Bastable, 1921), the thrust of which is as
follows: (i) Protection must be temporary and the infant industry should mature to the point where
it becomes viable without protection. (ii) The cumulative net benefits provided by the protected
industry should exceed the cumulative costs of protection. Krueger and Tuncer (1982), using
Turkish data for 1963-1976, is the first study to look for a correlation between infant-industry
protection and productivity growth; they concluded that the empirical evidence did not provide
support for the infant-industry argument. However, Harrison (1994), using the same data set,
showed that more protected sectors did in fact exhibit higher productivity growth. Bardhan (1971),
Redding (1999) and Melitz (2005) explored the conditions on the learning process under which
the benefits of protection justify the initial costs. Melitz (2005) also studied the optimal way in
which protection should be granted and Rodrik (2006) documented that much of China’s export
surge occurred simultaneously with the imposition of high tariffs during the period 1985-2004.
The third argument for IP is based on the idea that some industries exhibit Marshallian
externalities and/or inter-industry externalities, which are local externalities due to the existence of
transportation costs. These externalities can arise through labor pooling and input-output linkages
which, together with transportation costs, ensure that the pecuniary externalities remain local and
increase with the size of the industry (Marshall, 1920; Porter, 1990; Krugman, 1991; Fujita et al.,
1999). Well-known cases of Marshallian externalities are Silicon Valley, Route 128 and Italian
clusters, and the Manufacturing Belt in US, all of which have been emphasized in the literature on
the New Economic Geography (NEG) as examples of inter-industry externalities. Due to the
increasing returns to scale and transportation costs, the economy usually exhibits multiple
equilibria in the NEG models, which makes it reasonable for the government to use IP to achieve
an equilibrium which has a higher social welfare.
Finally, several recent studies focus on the argument of industrial policy for industrial
upgrading. Hausmann, Hwang and Rodrik (2007) argued that the varied economic performance of
different countries is partly explained by the goods that they produce. Harrison and
Rodríguez-Clare (2010) emphasized that there are “special industries”, and that countries can
increase welfare by reallocating resources to those industries. Other things being equal (including
physical and human capital stocks), countries that specialize in “strategic industries”, which
3
provide more opportunities for learning by doing, will trigger technological and industrial
upgrading. Within the framework of the new structural economics, Lin (2012) argued for a
proactive role for the government in nurturing development by actively supporting industries that
contribute to growth during industrial development and upgrading.
Our paper is related to the large body of empirical work on the effect of policy changes on
industrial and regional development, including specific applications to China. Governments of
many emerging-market countries, such as China, are more likely to play a proactive role in
directing financial resources than governments in developed countries. In the allocation of capital
they tend to favor state-owned firms and firms that have strong ties to the state (Song et al., 2011).
Economic zones are an important vehicle of IP in China and have recently been the subject of
several papers (Wei, 1993; Cheng and Kwan, 2000; Wang, 2013; Alder, 2016) studying their
impact on economic development. While these papers focus on the effectiveness and
heterogeneity of IP across places, ours is more interested in the effectiveness and heterogeneity of
IP across industries.
Our contribution to the literature is twofold. First we investigate the determinants of an
industry’s being selected in a PFYP, focusing particularly on the influence of the national
government’s selections in its NFYP but also analysing the importance of industry- and
province-specific characteristics. Second, we present an empirical analysis of the impact on
industry output of its selection in the FYP to assess the benefits of IP in China. More specifically,
we first estimate the predicted selection probability of an industry using a probit model. Then we
use difference-in-differences (DID) analysis to evaluate whether IP actually accelerates the growth
of the selected industries during the period covered by the FYP. Finally, we test whether the effect
of IP persists beyond the period covered by the FYP. Using Chinese data disaggregated to the
four-digit industrial level, we find that the central government’s preferences act as a key
determinant of industry selection by the provincial governments in drafting their FYPs, while,
surprisingly, industry productivity is not an important determinant of selection. Meanwhile,
preferential policy does promote output of the selected industries during the period covered by the
FYP but there is no evidence to support a persistent, long-lasting effect for industrial development
beyond the end of the FYP.
The rest of the paper is organized as follows. Section 2 presents a brief introduction to
China’s FYPs at national and provincial levels. Section 3 discusses data sources and data
processing methods used in the paper, defines key IP variables for empirical analysis in the
following two sections, and summarizes the main features of the 9th to the 12th NFYPs and PFYPs.
Section 4 presents the determinants of provincial industrial policy based on probit models. Section
5 further discusses the impact of policy intervention on output in the short and long term. Section
6 concludes.
2. Background: China’s Five-Year Plans
4
Since its establishment in 1949, the People’s Republic of China has released thirteen national
Five Year Plans (see Table 1 for the year of issue and the coverage period of each NFYP since
1949), providing a series of programs containing detailed economic development guidelines for all
its regions. As shown in Table 1, the First Plan (1953-1957) was formally approved by the
National Congress of the Communist Party of China in March 1955 and then officially adopted
and issued by the National People’s Congress in July of the same year. The key tasks highlighted
in the First Plan were to concentrate efforts on the construction of 694 large and medium-sized
industrial projects. Soon after the issue of the First NFYP the central government urged the
officials of provincial governments to draft their own five-year plans in the light of the First NFYP
and these were adopted one after another later in 1955. After China started market-oriented
economic reforms in 1978, there were gradual changes in the formulation and contents of NFYPs.
First, the central government launched NFYPs more regularly for every successive five years. In
particular, since the 9th FYP, the Proposal for the next NFYP is formally approved after the fifth
plenary session of the National Congress of the Communist Party of China held routinely in the
last November before a new five-year period begins. Then, the Outline for the new FYP is
officially approved by the National People’s Congress in March of the following year after
hearing opinions from all sides. The second change since 1978 is that, although setting
quantitative targets of development (e.g. the growth rates of GDP and income, the rate of
urbanization and the extent of industrial upgrading) are preserved as in earlier plans, principles
and instructions have gradually taken the place of particular construction projects in the NFYP.
However, more detailed targets of growth, projects and preferential guarantee conditions derived
in accordance with the targets and instructions of the NFYP are still to be found in special FYPs
for particular industries.1 The third innovation since 1978 is that PFYPs and special PFYPs for
particular industry are approved and issued by the provincial governments soon after the
announcement of the corresponding NFYP.
(Insert Table 1 around here)
Take the 12th NFYP (covering 2011 to 2015) as an example of a typical compiling process of
a FYP.2 The formulation of the 12th NFYP consisted of several steps, including: (a) preliminary
investigation (from 2008-2010); (b) formation of basic ideas and the drafting of the Proposal (in
the autumn of 2010); (c) approving the Proposal (in November, 2010); (d) preparing and
discussing the Outline by the National Planning Committee of Experts (NPC) (in the winter of
2010); (e) asking for internal and external suggestions (from 2010 to 2011); (f) approving the
Outline by the NPC (in March, 2011); and then (g) publishing and implementing the plan (during 1 Examples of special FYPs for particular industries in the 12th five-year round include the Industrial Upgrading Plan (2011-2015), the 12th FYP for New Material Industry and the 12th FYP for Sophisticated Equipment Manufacturing Industry. See the website of the Central People’s Government of the People’s Republic of China for more details at http://www.gov.cn/zwgk/2012-05/10/content_2104148.htm. 2 See Hu (2011) for a more detailed record of the compiling process of the 12th NFYP.
5
2011-2015). Since 1996, the formulation of the NFYP has become a public process with broad
participation, with the goals of the each NFYP being widely known before the plan is formally
approved. The formulation of the PFYP follows a similar procedure, except that planners would
already know the main features of the corresponding NFYP.
NFYPs and PFYPs are definitely not a matter of mere formality. First, they provide an
essential basis for drawing up more detailed industry-specific five-year plans, including
investment schedules by the provincial government and projects by local leading enterprises.
Second, they chart the course of regional policy for the next five years and therefore have a
significant impact on the province and the regions which interact with that province. Finally, the
requirement by the central government for a quantitative interim assessment of the implementation
of FYPs at various administrative levels since 2000 forces PFYPs to be executable.3 In short, the
FYP is one of the most significant indicators of the relationship between the central and local
governments and how regional governments take part in the reallocation of resources among
industries and influence the economic performance in the short run and in the long run
3. Data and stylized facts Data used in this paper include industrial policies extracted from the latest four NFYPs and
PFYPs (from the 9th to 12th, covering the period 1996 to 2015), as well as industry data for 31
Chinese provinces, autonomous regions and municipalities (hereafter abbreviated as provinces)
obtained from CSMAR’s China Industrial Statistics Database4. A provincial panel data set for
419 four-digit manufacturing industries for the period 1999 to 2010 was constructed to investigate
the questions of interest.
Based on NFYPs and PFYPs, we divide industries into two categories for each FYP period:
preferred industries and non-preferred industries. An industry is identified as preferred for a
certain FYP if in this FYP it is described as an “advancing industry”, a “pillar industry”, a
“promising industry”, a “priority industry”, a “breakthrough industry”, or it is planned “to enlarge
and strengthen”, “to prioritize”, “to develop”, “to accelerate”, “to expand” or “to cultivate” this
industry in the future when setting the goals for development for the next five years. The
remaining industries are collectively termed “non-preferred industries” and they are either not
mentioned in a FYP, or their prospects are designated as “to be rationally developed”, “to be
3 An interim assessment of the 10th NFYP (2001-2005) was first carried out by China’s National Development and Reform Commission (NDRC) in 2003, and such an assessment was soon followed by several east-coast provinces. Since then, the Law of the People's Republic of China on the Supervision of Standing Committees of People's Congresses at Various Levels was enacted in 2006, under which FYPs at various administrative levels are acquired by law for the first time to undergo an interim assessment and were made subject to supervision. An investigation conducted by NDRC on the interim assessment of the 11th PFYPs also demonstrated that the PFYPs have a substantial impact on local development. 4 The China Industrial Statistics Database is a sub-database of CSMAR (China Stock Market Trading Database) developed by GTA Education Technology Limited, which includes aggregated data for 39 two-digit industries and 714 four-digit industries. The industrial data in this database is based on firms that are “state-owned and above-scale non-state-owned manufacturing enterprises”, where “above-scale” requires the main business income (that is, sales) of an enterprise to be no less than 5 million RMB (revised to 20 million RMB in 2011). More information about this database is available from the Users’ Guide for China Industrial Statistics Database (version 2015) by CSMAR.
6
relatively controlled”, “to be optimized and adjusted”, “to be transformed”, “to be reduced in
scale”, “to be gradually eliminated”, “to be limited”, “to be orderly transferred” and so on.
Once an industry is selected as preferred in a PFYP, it will receive favorable support from the
provincial government in the next five years, including, but not limited to, priority in
construction-land usage, tax exemption, financial support from special fiscal funds of the
provincial government, expedited clearance of imported equipment and parts, deduction of
research expenditures before tax, double amortization of the cost of intangible assets, priority in
purchasing raw materials and electricity, free employee training programs offered by the
provincial government and so on.5 A policy dummy variable is introduced with a value of one for
preferred industries and zero otherwise.
Next, it is necessary to match the policy dummy to four-digit industry level data from
CSMAR. This is challenging since industries mentioned in FYPs and in CSMAR do not always
correspond. We use the following rules to solve this problem. First, if a preferred manufacturing
industry is mentioned in a FYP by a four-digit industry name or by representative products of a
four-digit industry, then this four-digit industry is identified as a preferred industry. Second, if the
preferred industry is mentioned by the name of a two-digit or three-digit industry, then each
four-digit industry belonging to this two-digit or three-digit industry is identified as a preferred
industry. Thirdly, if the preferred industry is mentioned by the collective name of a certain type of
industry, we refer to the specific guide on industrial classification and codes for statistics and
match it to the corresponding four-digit industry.6
A further complication is that the National Bureau of Statistics of China (NBSC) changed its
system of industry classification in 2002.7 To make the industry codes in the whole sample period
(1999–2010) consistent, we converted the four-digit industry codes in the 1999–2003 data to new
ones according to the correspondence table8 or by assigning a new code to a group of old codes
5 Take, for example, the new- and high-tech industry which has been emphasized since the 9th NFYPs. Firms identified as new- and high-tech firms are entitled to exemption from income tax for the first two years, a 10% reduction for the 3rd year, and weighted reduction of R&D expenditure pre-tax according to State Administration of Taxation. In Zhejiang province, which also prefers the new- and high-tech industry, cash grants of between 30 to 300 thousand RMB as well as various subsidies are offered to new- and high-tech firms in addition to the tax relief. More information about subsidies for the new- and high-tech industry in Zhejiang province can be found at http://www.96871.com.cn. 6 In a typical FYP, the manufacturing industries are mentioned either by representative products of a given four-digit industry, by name of the four-digit or two-digit industrial sectors (such as chemical industry), or by general name of certain type of industries (such as high-tech industry and new-material industry). We allocate each policy-preferred industry in the FYPs to the four-digit level by referring to various industrial classification standards established by the National Bureau of Statistics of China (NBSC) and supplementary guides for industrial classification provided by provincial bureaux of statistics, including the Code of Industrial Classification for National Economic Activities (GB/T4754-2002), the Product Classification for Statistics (NBSC, 2010), the Classification of Strategic Emerging Industry (NBSC, 2012), the Product Classification of New Materials, the Classification of High-tech (Manufacturing) Industry (NBSC, 2013), the Classification of High-tech Industry for Statistics (NBSC [2002] No. 33), the Interim Provisions for Classification of IT Industry for Statistics, the Classification of Environment Protection and Industrial classification for ocean industries and their related activities (GB/T 20794-2006).For more details of Industrial classification, see the website of the NBSC at http://www.stats.gov.cn/tjsj/tjbz/. 7 A new set of codes (GB/T 4754-2002) was adopted to replace the old one (GB/T 4754-1994) that had been used from 1995 to 2002. 8 In the case of a new four-digit code corresponding to an old four-digit code or several new four-digit codes corresponding to an old four-digit code.
7
based on product information9, as did Lu & Tao (2009). Given data limitations, all two-digit
manufacturing industries except the Waste resources & materials recovering industry are included
so that finally there are 419 four-digit manufacturing sectors in our analysis.
In the first part of our analysis, we not only examine the effect on industry selection in a
PFYP of selection in the NFYP but also consider a number of industry characteristics as additional
determinants, including an industry’s total factor productivity (TFP). To do so, we use data from
the Annual Survey of Industrial Firms (AFSI) for the years 2000 and 2005 from the NBSC to
calculate firm- and industry-level TFP. The AFSI data are cleaned and processed as suggested by
Nie et al. (2012); firm-level TFP is computed using the semiparametric method suggested by
Olley and Pakes (1996). Based on firm–level TFP, provincial industry-level TFP is calculated as
an output-weighted average of firm-level TFP.
Nominal variables are adjusted using price indexes taken from China Urban Life and Price
Yearbook (NBSC, various years).10
In the rest of this section we provide some statistical material, both as background
information to underpin later analysis and as an informal approach to some of the issues we deal
with more formally later in the paper. We begin with a comparison of the number of industries
listed as preferred in PFYPs and NFYPs for the latest four FYPs; these are presented in Table 2.
For provincial FYPs, on average there are 91 out of 419 (or 21.71%) four-digit manufacturing
industries that were selected for special development policy by at least one province. Interestingly,
the mean number of preferred industries per PFYP is smaller than in the corresponding NFYPs.
One possible reason is that the central government sets the course for the country as a whole and,
therefore, has more general goals of industrial development. Also, some provinces might not have
particular industries or have them but they are too small to be worth selecting.
(Insert Table 2 around here)
The above aggregate picture does not show the overlap between the provinces in their
selection of industries. We consider this in Table 3 which lists the 30 most frequently selected
industries (sorted by popularity in the 12th PFYP). The numbers in the last four columns are the
number of provinces which chose that industry in the relevant PFYP. Grouping industries by their
two-digit aggregates, it is clear that industry 27 (Pharmaceutical manufacturing), 37 (Auto
manufacturing) and 26 (Chemical raw materials & chemical products manufacturing) are the most
attractive two-digit industries for provincial planners. From Table 3, it can also be seen that, in the
12th PFYPs, of the most popular 30 industries, 22 to 30 of the 31 provinces simultaneously chose
9 In the case of several old four-digit codes corresponding to a new four-digit code. 10 The PPI and the GDP deflator are from China Urban Life and Price Yearbook. Since price indices for three two-digit industries are absent before year 2003, viz., the Agricultural products and byproducts processing industry, the Printing industry, and the Ordinary machinery manufacturing industry, this paper uses price indices for the Food manufacturing industry, the Paper and paper products industry, and the Special purposes equipment manufacturing industry of corresponding years to replace the missing deflator for the above industries before 2003.
8
these industries as their preferred ones, showing highly similar interests and judgements of
planners in different provinces. There is also considerable uniformity in the ranking of selected
industries across PYFPs; in Table 4 we report the relevant Spearman’s rank correlation
coefficients. They show considerable correlation across plans and, moreover, that the magnitude
of the correlation coefficient between adjacent plans increases over time, implying that the popular
industries are actually favored by more provinces in the next round. In other words, the selection
of preferred industries between provinces is becoming more and more similar.11
(Insert Table 3 around here)
(Insert Table 4 around here)
While there is a high level of consistency across provinces in the choice of the top 30, there is
still a considerable diversity in the number of industries chosen per province. This is illustrated in
Table 5 which lists the maximum, minimum and mean value of preferred industries selected by
provincial governments in the 9th to the 12th PFYPs. It is clear that the number of favored
industries varies considerably across provinces; the province which selected the maximum number
of favored industries chose almost eight times the number of industries chosen by the province
which selected the minimum number. If we consider the location of the provinces, it seems that
provinces in the east coast chose more than those in the central region which, in turn, chose more
than western provinces. This may be attributable to the more diversified industrial base of the
coastal and central provinces and/or the greater concentration in the coast and the center of the
type of industry typically chosen in the NFYP.
(Insert Table 5 around here)
We now look at the main features of the preferred and non-preferred industries at the time
when the next FYP was being drafted, i.e., in the years 2000, 2005 and 2010, or one year before
approval of the 10th, 11th, and 12th PFYPs respectively. Figures 1 to 6 illustrate the average
characteristics of the two groups of industries and it is clear that the preferred industries are indeed
quite different from the non-preferred ones.
We begin, in Figure 1, by comparing average firm size in the preferred and non-preferred
industries, where firm size is measured by the average number of employees per firm in the two
types of industry. It is clear from Figure 1 that preferred industries on average have firms which
are about 40% larger than those in the non-preferred industries. This finding is consistent across
11 The ranking of preferred industry is generally highly consistent through time except for industries within the two-digit Chemical raw materials & chemical products manufacturing industry (with two-digit code 26). The fluctuation in the number of provinces which selected these industries in the 11th PFYPs is noticeable and could be associated with a sudden reduction of 25% in numbers of preferred industries in this two-digit industry in the 11th NFYP, compared to the other NFYPs. The selection of PFYP are highly correlated with that of the NFYP; we provide evidence for this relationship between NFYPs and PFYPs in Table 6 and in section 4.
9
the three years considered with the largest differences being in 2005.
We turn, in the remaining figures to measures of industry size. In Figure 2 we picture
industry size as measured by the average number of employees per industry for preferred and
non-preferred industries. The data show that policy-preferred industries have more employees. In
2000, the mean number of employees in the preferred industries was more than 2 times larger than
that of non-preferred industries; the ratio increased to approximately 2.5 by 2010.
Besides number of employees, Figures 3 to 5 indicate that, on average, preferred industries
produce more gross output, more value-added and also have much larger net value of fixed
assets.12 In 2000, the magnitude of gross output, value-added and fixed assets of preferred
industries were 2.9, 3.1, and 2.2 times larger than that of the non-preferred industries. And it is
clear that the differences between preferred and non-preferred industry groups remain stable
through time.
(Insert Figure 1. to Figure 6. around here)
A surprising contrast is the productivity difference between the two types of industry. Figure
6 shows the average industry level of TFP for preferred, non-preferred and total industries; as
indicated earlier, industry TFP was calculated from the weighted average of firm-level TFP with
firm’s gross output as weights. It is puzzling that the mean value of preferred industries’ TFP in
Figure 6 is slightly lower than that of the non-preferred ones. Two possible explanations suggest
themselves. One is that the government does not really care about the benefits of preferential
treatment in terms of overall economic growth but wants to benefit relevant interest groups which
are concentrated in large industries – they are more important (say, large industries dominated by
state-owned enterprises which happened to have lower TFP).13 The other possibility is that the
provinces intentionally pick (some) low TFP industries because they see potential for TFP growth
in them. Further, more formal, exploration of the relationship between selection and productivity
growth will be left to a later section.
Finally in this section, we turn to a preliminary examination of the relationship between
NFYPs and PFYPs. Table 6 presents evidence that a provincial government is more likely to
select an industry for policy-preference if it has already been chosen for preference by the national
government. Among all the four-digit manufacturing industries, industries preferred by the central
planners on average attracted more provinces than those not preferred by the central government.
Thus in the 9th FYP cycle an industry already chosen by the national government was chosen by
11.5 provinces on average while one that had not been selected by the national government was 12 Note that data for 2010 are not included in Figure 5 since data for value-added for 2010 are absent from the CSMAR database. 13 Among numerous papers which analyze TFP for Chinese firms, one of the common findings is that State-Owned Enterprises perform more poorly than Domestic-Private Enterprises. From 1992-2007, TFP in the private sector is believed to have been about 80 percent higher, on average, than in the state sector (Curtis, 2016). Moreover, TFP growth in industrial SOEs by the mid-1990s may even have turned negative (Laurenceson and Chai, 2000; Jefferson, Rawshi, Wang and Zheng, 2000). See also Storesletten and Zilibotti (2014).
10
chosen by only 6.8 provinces on average. This selection pattern was more pronounced with each
successive FYP. The last column of Table 6 reports a t-statistic for a test of significant
difference between the number of industries chosen in the 9th and 12th plans and shows that the
number of provinces following the national government’s choice significantly increased while the
number choosing industries not chosen by the central government fell significantly; the overall
number selected did not change significantly over the period. Thus, provincial planners focus
more and more on the industries preferred by the central government.
(Insert Table 6 around here)
In summary, the above stylized facts imply that provincial planners are inclined to accelerate
the development of industries which, on average, have larger firms, more employees, more fixed
assets, greater output, higher value-added, and those favored by the central government. However,
industry TFP does not appear to be an important determinant of selection by provincial planners;
if anything, planners choose industries with lower average TFP. We go on in section 4 to assess
more formally the impact of these factors on the choices made by provincial planners.
4. Provincial governments’ choice of industrial policy In this section we report the first part of our empirical analysis in which we address the
question of how industry policy is formulated in China. In particular, we examine the response of
provinces in their choice of preferred industries to the selection of preferred industries by the
central government before examining the importance of a number of industry characteristics for
the selection process. Recalling the discussion in section 2 that PFYPs are drafted when the
corresponding NFYP is already publicly known, a positive impact of industrial selection in NFYP
on PFYPs is expected.
As discussed in Section 3, an industry may enjoy various benefits from selection: subsidies
from government which may be financial, such as tax rebates and R&D subsidies, or policy-based,
such as land preferential policies (see Hua et al., 2016; Lee et al., 2014). These Government
subsidies are usually intended to achieve a number of social and political objectives including
promoting export competiveness, production efficiency, employment, and social welfare (Lee et
al., 2014). Our earlier informal analysis also showed that policy-preferred industries are quite
different from non-preferred industries. We now move to more formal analysis of the choice by
provincial governments of their favorite industries and we do so by estimating the predicted
probability of an industry’s being selected using a probit model of the form:
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖∗ = 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃(𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖 = 1|𝑍𝑍) = 𝑍𝑍′𝛽𝛽 + 𝑢𝑢 (1)
where i indexes industry, r indexes province, and n indexes period. The dependent variable,
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖∗ , is the probability of industry i being chosen by province r as a preferred one; it is
dependent on 𝑍𝑍, a series of social and economic characteristics of industry i and province r.
11
𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖 is a binary variable, identifying industry i as selected or not (1, 0 respectively) by province
r in its nth PFYP. In section 3, when discussing Figures 1 to 5 we showed that the preferred
industries have several characteristics, viz., more employees, more output, more value-added, and
a higher probability of having been favored by the central government. In practice, these
characteristics may be meaningful, because supporting large industries and maintaining a high
employment level can be important for subsequent economic growth and local stability. As for the
relationship between the NFYP and PFYPs, it may be argued that following the central
government’s decision about preferred industries not only shows the provincial planner’s loyalty,
but also gains extra resources from the central government for the purpose of local development In
addition, both loyalty and better economic performance are likely to benefit the provincial
official’s political career in a so called “promotion tournament” (Zhou, 2005). Therefore, we use
an empirical model with explanatory variables which include the central government’s preferences
regarding favored industries, the number of industry employees, industry gross output and
industry taxes and fees paid in province r. More specifically, the probit model is specified as
follows:
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃(𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖 = 1|𝑍𝑍) = 𝛽𝛽0 + ∑ 𝛽𝛽𝑗𝑗 . 𝑃𝑃�𝑃𝑃𝐶𝐶𝑃𝑃𝑃𝑃𝐶𝐶𝑖𝑖𝑖𝑖𝑗𝑗+1 = 𝑗𝑗 + 1�3
j=1 + 𝛽𝛽4𝐶𝐶𝑇𝑇𝑇𝑇𝑇𝑇𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖𝑖𝑖 + 𝛽𝛽5𝐶𝐶𝑀𝑀𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖 +
𝛽𝛽6𝑃𝑃𝑃𝑃𝑂𝑂𝑂𝑂𝑇𝑇𝑂𝑂𝐶𝐶𝑖𝑖𝑖𝑖𝑖𝑖 +∑ 𝑃𝑃𝐶𝐶𝑃𝑃𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘 𝛽𝛽𝑘𝑘𝐾𝐾𝑘𝑘=7 + 𝑢𝑢𝑖𝑖 (2)
where CTYPEin is a categorical variable that captures whether industry i has been chosen by the
central government in the previous two NFYPs. To be specific, CTYPEin equals to 1, 2, 3, and 4
respectively if industry i was never preferred in the latest two NFYPs, only preferred in the
last-but-one NFYP, only preferred in the last NFYP and preferred in both NFYPs since the nth
NFYP (see Table 7 for details).
(Insert Table 7 about here)
Then 𝑃𝑃�𝑃𝑃𝐶𝐶𝑃𝑃𝑃𝑃𝐶𝐶𝑖𝑖𝑖𝑖𝑗𝑗+1 = 𝑗𝑗 + 1�, 𝑗𝑗 ∈ (1, 2, 3), is an indicator operator which generates dummy
variables for three of the four cases above. Furthermore, we choose the number of industry
employees (EMPirn), industry total tax and fee payment (TAXFEEirn), industry output share
(OPSHAREirn) and industry productivity (TFP) as core explanatory variables. EMPirn is defined as
the total number of employees belonging to industry i in province r one year before the nth PFYP
is drafted, TAXFEEirn includes total tax and extra charges on sales of products plus income tax of
that industry in the year before the nth PFYP, and OPSHAREirn is gross output value of industry i
in province r divided by total gross output of province r. All of these independent variables refer
to the year before the nth PFYP drafts are released. Finally, 𝑃𝑃𝐶𝐶𝑃𝑃𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖𝑖𝑖𝑘𝑘 represents a set of dummy
variables including period dummies, industry dummies and province dummies, with 𝛽𝛽𝑘𝑘 the
corresponding regression coefficients.
The baseline estimation results are listed in Table 8. In the first column of Table 8, we
include only policy variables and a set of dummy variables. The sign and magnitude of estimated
12
coefficients of the CTYPEin variables indicate that the central government’s NFYPs have a
profound effect on provincial policy design. The coefficients are all positive and significant,
meaning that no matter whether an industry was selected once or twice in the most recent two
NFYPs, there is a significant increase in its predicted selection probability by the provincial
governments. In addition, the relative magnitudes of the CTYPEin coefficients also seem plausible:
the probability of selection by a provincial government is higher if the industry was selected by
the central government in the most recent NFYP than in the one before and, furthermore, the
selection probability is higher still if the industry was selected in both the most recent NFYPs.
These results are completely unaffected by the addition of other influential variables: the
number of employees, taxes & fees and the industry’s output share. Neither the significance nor
the relative magnitude of the estimated policy effects changes when variables are added and,
indeed, the magnitudes of the coefficients themselves hardly changes, indicating that the impact of
the NFYP is quite robust. At the same time, the sign of the coefficients of other control variables
are also significantly positive and consistent with our expectations: industries having a larger
number of employees, paying higher taxes and fees, or contributing more to provincial output tend
to have a higher predicted probability of being selected by the provincial government.
(Insert Table 8 around here)
(Insert Table 9 around here)
In Table 9, we report the predicted probability of CTYPEin for industry i, when holding all
other variables in column 5 of Table 8 at their means. Other things being equal, if an industry is
never selected in the latest two NFYPs, the predicted probability of its selection by PFYP is 9.85%.
If an industry is preferred in the last-but-one or in the latest NFYP, the predicted probabilities
raise to 16.24% and 21.08% respectively. Although being selected in the most recent NFYP has a
larger impact on the predicted probability of selection than being selected in the last-but-one, a
t-test shows the difference is not statistically significant, i.e., the coefficient of 𝑃𝑃𝐶𝐶𝑃𝑃𝑃𝑃𝐶𝐶𝑖𝑖𝑖𝑖3 is not
significantly larger than that of 𝑃𝑃𝐶𝐶𝑃𝑃𝑃𝑃𝐶𝐶𝑖𝑖𝑖𝑖2 . However, when an industry is selected in both the latest
two NFYPs, the predicted probability increases dramatically to 51.61%, which is significantly
larger than all the former ones. Therefore it can be concluded that an industry’s probability of
selection by the provincial planner increases monotonically with the number of times an industry
is selected by the central planner and if the policy enjoyed by an industry changes from not
favored in the latest two NFYPs to selected in both NFYPs, the probability of its being selected as
a policy-preferential industry by the provincial planner rises significantly by 41.76 percentage
points. Table 9 also reports the estimated marginal effects of the main other independent variables.
If the number of total employees, taxes and fees or output share of an industry go up by an
infinitesimal amount, the probability of its being selected as a preferred industry by the provincial
planner rises significantly by 0.25%, 1.54% and 4.00% respectively.
13
To investigate the relation between NFYP and PFYP more thoroughly, we consider whether
the region in which the province is located has a systematic effect on the selectin probability. We
distinguish between two regions: inland and coast and add interaction terms between region and
the CTYPEin variables to the baseline specification in which inland equals 0 for the eastern coastal
provinces.14 It is well known that the inland provinces generally lag in their development
compared to the coastal provinces. The results are reported in Table 10.
(Insert Table 10 around here)
Table 10 shows that the sign of the interaction term is consistently positive no matter which
other variables are included in the equation, suggesting that the inland provinces are inclined to
follow the central government’s selection more closely than the coastal provinces are. More
particularly, the magnitude of the CTYPEin coefficient for the inland provinces is much larger
when an industry is preferred in the latest NFYP. Since the inland provinces are more dependent
on resources obtained from the central government, it is not surprising that they have a greater
enthusiasm for following the industrial selection by the central government. This confirms the
finding of Wu and Zhu (2015) that inland provinces are likely to adapt their industrial plans much
more readily to that of the central government than coastal provinces are.
Finally we consider whether an industry’s productivity influences positively its predicted
probability of selection by a regional government. A high TFP level in the recent past may
generate optimistic expectations that an industry will growth rapidly in the future and is therefore
worth selecting for preferential treatment. Therefore, in this part we focus on whether the level of
an industry’s TFP is another determining factor in the choice by provincial planners. In particular,
are they inclined to select industries with a high level productivity growth (as a predictor of more
high productivity growth in the future)? Following De Loecker (2007), we estimate firm-level
TFP using the framework of Olley and Pakes (1996), where the export decision is introduced as
dummy variable.15 For industry i in province r, TFPirn is calculated as the weighted average of
firm-level TFP one year before the nth FYP is approved (using firm-level output as weights). We
first use TFPirn as a regressor in the probit selection equations. The results are reported in Table
11.16 Next, we use dummy variables to pick out the top 10% and top 20% of industries according
to their TFP. Recall from Table 1 that, on average, PFYPs select about 20% of industries. In
light of this, two dummy variables, H-TFP90 and H-TFP80, are introduced to identify high-TFP
industry groups. To construct these two dummies, we first sort industries according to TFPirn in
each province, and set H-TFP90 equal to 1 for industries with their TFP levels ranking in the top
0.10 quantile in that province and 0 for the others. Similarly, H-TFP80 is set equal to 1 for
14 The eastern coastal provinces include Beijing, Tianjin, Hebei, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, and Guangdong province. 15 Due to value added of year 2010 is absent in ASIF, analysis of TFP focus only on year 2000 and 2005. 16 We also use simple average of firm level TFP as a proxy, and the results remain.
14
industries with their TFP levels ranking in the top 0.20 quantile in that province and 0 for others.
The results of the use of these dummy variables in the probit equations are reported in Table 12.
(Insert Table 11 around here)
(Insert Table 12 around here)
To our surprise, the coefficient of TFPirn in column (1) of Table 11 is significantly negative,
meaning higher productivity doesn’t imply a higher chance of selection by the provincial
government. When we further control for other variables in columns (2) to (5), this result remains
unaffected – the coefficient is consistently negative and significant although the absolute
magnitude falls somewhat suggesting possible negative collinearity with industry size – the larger
the industry the lower the TFP . The results in Table 12 are consistent with those in Table 11 – the
coefficient of the dummy variables is invariably negative although only H-TFP80 is significant.
Whether we look at the top 20% or at the top 10% of the most efficient industries, a higher TFP
level does not improve these industries’ chance of selection, whether we control for taxes and fees
or output share or not. We can conclude that provincial planners selected industries without
considering whether the chosen industries are among the most efficient industries in that province.
As is well known, TFP is an important source of long-run growth so that a natural question is why
planners don’t choose to promote the most productive industries? One possible reason is that the
PFYP usually sets a short-run growth target, which can be achieved with less risk by taking
advantage of the traditional large industries, even though the productivity of these industries may
be inferior to that of some emerging industries. This may be rationalized on the basis that an
industry with high a TFP level but limited current scale cannot contribute greatly to the growth of
provincial gross output or fiscal revenues in the short run. In addition, traditional industries might
also be supported because of their contribution to maintaining local employment or having better
relations with the provincial government.
All in all, we can conclude from the above analysis that the central government’s preferences
clearly act as a key determinant of provincial governments’ selection of preferred industries, while
industry output scale, taxes and fee paid, and employment also play significant roles. However
high productivity doesn’t improve an industry’s chance of selection by the provincial government,
which raises suspicions about the impact of policy on industrial performance. In section 5 we will
test the effect of industry selection in PFYPs on industrial outcomes in the short run and in the
long run.
5. The effect of industrial policy on output We now turn to the second question of interest in this paper: do the selected industries
develop faster because of the policy? And if the answer is “yes”, then how long is the effect
maintained? To address these issues we proceed as follows. First, we will use the
15
difference-in-differences (DID) method to measure the average treatment effect on the preferred
industries. Second, to accommodate possible endogeneity in policy selection, we will also use a
two-stage least squares method to estimate the effects on industrial outcomes (e.g., output growth,
fixed assets, and employment) of being selected in a PFYP. However, the impact of policy during
the currency of the PFYP captures only the short-term effect of policy and it is interesting to see
whether any beneficial policy effect persists beyond the term of the FYP even after preferential
support from the government ceases. Therefore in the third part of section 5 the propensity score
matching (PSM) method is combined with DID to analyze whether any beneficial effect lasts
beyond the current period of the relevant PFYP.17
5.1 Difference-in-differences analysis: basic specification
As mentioned before, China’s industrial classification underwent a major revision in the late
1990s, making data before and after 1998 incompatible. We therefore have available repeated
cross-section data sets for 2000, 2005 and 2010 which we can use for pre- and post-treatment data
for the 10th PFYPS (2001-2005) and the 11th PFYPs (2006-2010). In keeping with the standard
DID method, we take the preferred industries in the 10th or 11th PFYPs as the treatment group and
those not exposed to provincial preferential policy as the control group. In the standard two-period
DID specification where the same units within a group are observed in each time period, the
average gain in the control group is subtracted from the average gain in the treatment group. This
removes biases in second-period comparisons between the two groups that could be the result of
permanent differences between them, as well as biases possibly resulting from comparisons over
time in the treatment group that could be the result of trends. When focusing on the policy effect
of industry selection in the 10th and 11th PFYPs, we have three periods for each sample and the
specification of the multi-stage DID regression equation is as follows.
lnY𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 = α+ ∑ 𝛾𝛾𝑗𝑗 . 1(𝑛𝑛 = 8 + 𝑗𝑗)3𝑗𝑗=1 + ∑ 𝜃𝜃𝑗𝑗 . 1(𝑔𝑔 = 𝜏𝜏)𝐼𝐼
𝜏𝜏=1 + 𝛿𝛿𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖10&11 + ∑𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝜑𝜑 + 𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖 (3)
where i indexes the industry, r indexes the province, n indexes the period, and g indexes the group.
lnY𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 is the logarithm of industry i’s output in province r in the last year of the nth PFYP (n=9,
10, 11). The two indicator functions, 1(𝑛𝑛 = 8 + 𝑗𝑗) and 1(𝑔𝑔 = 𝜏𝜏), are equal to 1 if 𝑛𝑛 = 8 + 𝑗𝑗
and 𝑔𝑔 = 𝜏𝜏 respectively and 0 otherwise. 𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖10&11 is a binary policy variable similar to 𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖 in
Section 4 except that is focusses only on the 10th and 11th PFYPs. Finally, variables 𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 are
other covariates including industry dummies and province dummies. The entire sample is divide
into three groups, viz., the control group which includes industries neither preferred in the 10th nor
in the11th PFYPs, treatment group I (denoted Treat I) which includes industries selected twice and
treatment group II ( denoted Treat II) which includes those only selected once ever since the 10th
PFYPs. Thus, we can alleviate other uncontrolled influences and test robustness of the results
through the use of proper treatment groups.
17 In addition, we also use a two-stage panel model to assess the robustness of our estimates of the impact of industrial policies in the 10th (2001-2005) and 11th (2006-2010) PFYPs. The overall conclusions are similar.
16
(Insert Table 13 around here)
The results using the standard fixed-effects estimator for the multi-period model are reported
in Table 13. As show in column (1), the coefficient of 𝑃𝑃𝑇𝑇𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖10&11 is significantly positive,
meaning preferential policy has a stimulating effect on industry output. This result remains
consistent even if we drop from the sample those observations for which output ranks in the top 5%
or bottom 5% in each province to reduce possible bias caused by outliers; these are reported in
columns (2) and (3). However, the estimated coefficient of policy could still be biased because
industries in Treat I had enjoyed preferential policy in two successive five-year periods and the
impact of policy might be diminishing or increasing at the margin. For that reason, we re-run the
fixed-effects model excluding samples in the Treat I group and report the result in column (4). The
result remains: the estimated coefficient of provincial policy is significant and positive as formerly.
Additionally, we also consider possible uncontrolled bias that might result from a carry-over effect
of policy from the 9th PFYP (1996-2000) on industrial output in 2005 and 2010. To avoid the
possible influence of the 9th PFYP we further drop all the observations for industries which have
been selected in the 9th PFYP. At the same time, to better control group effects we split Treat II
into two parts according to the timing of policy as well: treatment groups II-I (denoted Treat II-I)
and II-II (denoted Treat II-II) are introduced for industries selected only in the 10th or 11th PFYPs,
respectively. The results are shown in column (5) of Table 13. Although we take into
consideration possible influences from the 9th PFYP and extra group effects due to program timing,
the estimated coefficient of preferential policy remains significantly positive and the magnitude
changes little from the previous specification, which confirms that industrial policy is effective
during the period to which the PFYP applies.
Despite the strong and robust results reported in Table 13, data limitations due to the change
in industrial classification in 1998, makes it impossible to enter unit-specific variables which
might also influence industry development into the equation; variables such as those which
capture the inter-firm competition effect and Marshallian externality. Moreover, we have not been
able to test the parallel-trend assumption which underlies the DID method. Finally, it is possible
that there is bias due to the endogeneity of policy selection, i.e., policy selection might be related
to industrial outcome at the baseline. We will deal with all these problems in section 5.2 but to do
so we will focus only on the effects of policy in the 11th PFYP for which a more comprehensive
data set at the four-digit industry level is available.
5.2 Difference-in-differences analysis: robustness test
In this section, we deal with problems that might cause biased estimates of the policy effect
and, in so doing, test the robustness of the results in Section 5.1. In particular, we analyze the
following four additional cases. First, we employ the baseline DID specification set out in
Equation (3) to estimate the impact of the 11th PFYP (2006-2010) with extra unit-specific
covariates as controls. Second, given the possible failure of the parallel-trend assumption which
17
underlies the standard DID model, the change of industrial output is used instead of output as the
dependent variable, which relieves the need for paralleled trends between the control group and
the treatment group. Third, since preferential policies are implemented by successive five-year
plans, some industries could be selected twice in adjacent plans. If policy has a persistent impact
into the next round, the estimated coefficient could be biased. For that reason, we re-run
regressions based on samples which omit industries affected in this way. Finally, as pointed out in
Section 3, ex-ante, average firm employment, industry fixed assets, industry gross output, and
industry value-added for the preferred industries consistently exceed those of the non-preferred
industries, so that it is possible that policy selection might be related to outcomes at the baseline.
To control for this possible source of endogeneity, we also implement a two-stage least squares as
well as an instrumental-variables approach to further test the robustness of the previous results.
For simplicity, we rewrite Equation (3) in form of the standard two-period DID model as
follows:
lnY𝑖𝑖𝑖𝑖𝑇𝑇𝑇𝑇 = 𝛿𝛿1 + 𝛿𝛿2𝑃𝑃𝑃𝑃 + 𝛿𝛿3𝐶𝐶𝑇𝑇 + 𝛿𝛿4𝑃𝑃𝑃𝑃 ∗ 𝐶𝐶𝑇𝑇 + ∑ 𝑃𝑃𝐶𝐶𝑃𝑃𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖𝑘𝑘 𝛿𝛿𝑗𝑗𝐽𝐽𝑗𝑗=5 + u𝑖𝑖𝑖𝑖 (4)
where i indexes the industry, r indexes the province, n indexes the period, and g indexes the group.
lnY𝑖𝑖𝑖𝑖𝑇𝑇𝑇𝑇 is the logarithm of industry i’s output in province r before and at the last year of the 11th
PFYP (i.e., 2005 if 𝐶𝐶𝑇𝑇 = 0 and 2010 if 𝐶𝐶𝑇𝑇 = 1). 𝑃𝑃𝑃𝑃 is a binary provincial policy variable
which equals one if the industry is preferred in the 11th PFYP and zero if not. 𝐶𝐶𝑇𝑇 is a time
dummy that equals zero for 2005 (before the 11th PFYP) and one for 2010 (after the 11th PFYP).
𝑃𝑃𝐶𝐶𝑃𝑃𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖𝑘𝑘 are other control variables including the lagged number of firms, 𝑇𝑇𝑃𝑃𝑂𝑂𝑀𝑀, and its square,
𝑇𝑇𝑃𝑃𝑂𝑂𝑀𝑀𝐶𝐶𝐹𝐹, which are used to control inter-firm Marshallian externality and competition effects,
following Belleflamme et al. (2000), Crozet et al. (2004), and Head (1995). In addition, the
controls also include a set of dummy variables to control location- and industry-specific influences.
In this form of the model, we are interested in estimating the coefficient of the interaction term, 𝛿𝛿4,
or the average treatment effect on the treated industries.
(Insert Table 14 near here)
The estimated coefficients for regression equation (4) are reported in the first three columns
of Table 14. They show a consistent, significant positive association between policy and gross
output and, also, the coefficients of the interaction term in columns (1) to (3) remain stable when
control variables are added, confirming the results in section 5.1. In addition, the signs of the
estimated coefficients of externality and competition effects are consistent with the predictions of
the theoretical model in Belleflamme et al. (2000) and Crozet et al. (2004), i.e., the Marshallian
externality effect shows a positive association with industrial development while the competition
effect, on the contrary, reduces growth. Finally, the estimated coefficients of the dummy variable
for location show that output growth is faster in the coastal provinces than in the inland provinces.
In the light of our conclusions in section 4, this looks reasonable since we found there that the
18
inland provinces have a higher likelihood of mimicking the central government’s choice which
might result in sacrificing their own comparative advantage.
One of the most common problems with DID estimates is the failure of the parallel-trend
assumption (Angrist and Krueger, 1999). Since we are focusing on the 11th FYP, we have two
periods before the treatment (2000 and 2005) and one period after the treatment (2010). So we can
use the conditional parallel-trend assumption and the corresponding regression model proposed by
Mora and Reggio (2012), which is available for applications with more than one pre-treatment
period and assumes that the average change in output for the treated industries had they been
untreated is equal to the observable average change in output for the comparable controls.
Obviously, the conditional parallel-trend assumption will hold if the unconditional parallel-trend
assumption holds, but not vice versa. This involves using the difference of industrial output so that
the traditional parallel-trend assumption is relaxed to obtain the following model.
ln∆Y𝑖𝑖𝑖𝑖𝑇𝑇𝑇𝑇 = 𝛿𝛿1 + 𝛿𝛿2𝑃𝑃𝑃𝑃 + 𝛿𝛿3𝐶𝐶𝑇𝑇 + 𝛿𝛿4𝑃𝑃𝑃𝑃 ∗ 𝐶𝐶𝑇𝑇 +∑ 𝑃𝑃𝐶𝐶𝑃𝑃𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖𝑘𝑘 𝛿𝛿𝑗𝑗𝐽𝐽𝑗𝑗=5 + u𝑖𝑖𝑖𝑖 (5)
where for industry i in province r, ln∆Y𝑖𝑖𝑖𝑖𝑇𝑇𝑇𝑇 is the logarithm of the output difference five years
before and after implemented of the 11th PFYP (i.e. 2001 to 2005 and 2006 to 2010) and notation
for other variables is the same as in equation (4). Again, we are interested in estimating 𝛿𝛿4, the
coefficient of the interaction term. The estimation results are listed in columns (4)-(6) of Table 14
from which it can be seem that the signs of the interaction term of policy and period are still
consistently significantly positive, no matter whether additional control variables are used or not.
Note that the estimated coefficient of the interaction term in column (4)-(6) is slightly larger than
those in column (1)-(3). One possible explanation for this is the non-efficiency-oriented process of
picking preferred industries discussed in Section 4, causing a downward bias in the estimate of the
effect of policy. More specifically, since some inefficient industries with lower growth potential
are singled out by the provincial government because of its willingness to copy the central
government’s choice or its aim to support friends, the coefficients of the interaction term increase
in regression models based on the conditional parallel-trend assumption. However, it is
noteworthy that, as shown in Table 14, this underestimation does not change the result that policy
has a significant and positive impact on industrial output during its current period.
Another source of potential bias is due to the possible persistent impact of policy in the sense
that being selected in one plan has effects beyond the end of the plan. In that case it would bias the
estimator due to some industries having been selected in a previous plan but the effect being
attributed to the current plan. To address this possibility we use an approach employed previously
and run the regression models in equations (4) and (5) with samples which are restricted to those
industries not picked in the 10th PFYPs; the results are reported in columns (1)-(4) of Table 15.
The parameter of core interest is still the coefficient of the interaction term. It can be seen in the
first two columns of Table 15, for which the logarithm of industry i’s output is used as the
dependent variable in Equation (4), the estimated coefficients of the interaction term are positive
and significant and similar in magnitude to those in Table 14. Moreover, the results are little
19
changed even if we take the logarithm of output difference as the dependent variable as specified
in Equation (5), estimates for which are reported in the third and fourth columns, further
confirming the robustness and plausibility of the estimation results.
Finally, we found in section 3 that informal evidence suggests that average firm scale, fixed
assets, gross output, and value-added of the provincial preferred industries are considerably larger
than those of the non-preferred industries, suggesting that policy selection might be related to
outcomes at the baseline. For this reason, the estimated coefficient of interest might suffer from an
endogeneity problem since if industries with lower (higher) growth potential are chosen
intentionally, it would lead to the underestimation (overestimation) of the policy impact. To assess
this possibility we employ a two-stage least squares approach together with instrumental-variables
(IV). To employ this method we need to find a valid instrument. Recall from the analysis in
Section 4 that the central government’s preferences have a strong influence on provincial
governments’ policy selection. Further, if the central government’s preferences were influenced
by industry characteristics they would relate to the performance of industry i in the country as a
whole, but not to the performance of industry i in any particular province which may influence
that province’s selection. Thus the central government’s policy selection could be employed as an
instrumental variable here. The instrument is constructed as shown in Table 16. To take advantage
of as much information as possible, we divide policy intensity into eight different types based on
the 9th-11th NFYPs, using definitions very similar to those of CTYPEin in Table 7. More
specifically, according to policy status in the 9th to 11th NFYP, a value 1to 8 is assigned to the
variable Ctype2 for each industry i as explained in Table 16. The higher the value of Ctype2, the
more important industry i is in the view of the central government. Since local governments have
a strong incentive to follow the central government’s policy, Ctype2 correlates highly with local
governments’ policy selection, as we found in the analysis reported in section 4. We then
estimated the following equation on cross-section data for the 11th PFYP.
ln∆Y𝑖𝑖𝑖𝑖𝑖𝑖 = 𝛿𝛿1 + 𝛿𝛿2𝑃𝑃(𝑃𝑃𝑃𝑃 = 1) + ln∆Y𝑖𝑖𝑖𝑖(𝑖𝑖−1) + ∑ 𝑃𝑃𝐶𝐶𝑃𝑃𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖𝑚𝑚𝛿𝛿𝑚𝑚𝑀𝑀𝑚𝑚=3 + 𝑢𝑢𝑖𝑖𝑖𝑖𝑖𝑖 (6)
where the notation in equation (6) is the same as that in Equation (5), the lagged dependent
variable has been added to the control variables as suggested by Wooldridge (2006) to relieve the
problem of omitted variables and the instrument explained above has been used for the
policy-selection variable. As usual, the estimate of the policy parameter 𝛿𝛿2 is the coefficient of
core interest and the estimated results of Equation (6) are reported in the last two columns of
Table 15. In concordance with our expectation, the coefficient of PFYP is consistently significant
and positive with a magnitude similar to previous estimates, once again supporting the hypothesis
that provincial industry policy stimulates the industry’s output level during the period of the plan.
Tests for the validity of IV, including the first-stage robust F-statistic, Shea’s partial R2, and
Wooldridge's (1995) robust score test (a test of endogeneity), are also reported in Table 15 and all
of them exceed the relevant critical value, thus ensuring the validity of IV. We also used GMM in
the event that there is heteroscedasticity. The results are not reported but the estimation results
20
are consistent with those in Table 15.
(Insert Table 16 near here)
We can summarize the results reported in sections 5.1 and 5.2 by pointing to the remarkable
robustness of the estimates (sign, significance and magnitude) of the crucial parameter measuring
the effects of provincial policy selection on an industry’s output in the period during which the
relevant PFYP is current. This adds to the findings of Song and Wang (2013) where the industry
policies in PFYPs is found to have a significant effect on productivity of the provincial pillar
industries. While in our paper, we test the effectiveness of industry policy in PFYPs on output for
all the preferred industries besides the pillar ones. Chen et al. (2016) shows evidence that
export-promoting policies expanded export of targeted industries in export processing zones,
which is a more specific example consistent with the results in ours. It seems that preferential
policy programs could indeed have a positive impact on development of the affected industry.
Our findings however, leave open the question of whether industrial policy has a persistent
effect on preferred industries beyond the currency of the PFYP in which the industry was selected.
We turn to this question in section 5.3.
5.3 Policy effects in the short and long terms
So far we have analyzed the contemporaneous effect on industry output of being selected as a
provincial preferred industry but we are also interested in whether the beneficial effect found in
the previous section persists over time. For that reason, we choose industries that were selected as
preferred in the 10th PFYP but not in the 11th PFYP as the treatment group, and industries not
selected in either the 10th PFYP or the 11th PFYPs as the control group. Using this sub-sample, we
first test whether policy interventions increased output of the treatment group during the period of
the 10th PFYP (2001-2005) and then evaluate whether there was any carry-over effect in the
subsequent five years after the preferential policy was cancelled, i.e., we re-evaluate the
performance of the treated group at end of the 11th PFYP period. Since local government doesn’t
select/discard preferred industries randomly, preferential policy could be cancelled because of
poor performance in particular industry. As discussed in section 4, if the difference between the
treatment group and the control group at the baseline correlates with policy selection, it will lead
to a biased estimator. Therefore, we employ the propensity-score matching (PSM) method
proposed by Rosenbaum and Robin (1983) to screen industries in the control group to evaluate the
net influence of policy intervention (the average treatment effect of the treated, hereafter ATT).
With the control group singled out by the PSM method, DID analysis can be used to eliminate the
influence of unobservable variables. It is found in this section that, although policy increases
output of a preferred industry during the current period of the plan in which it is selected, the
expansionary effect disappears once the preferential policy is cancelled, implying that temporary
invention by the government does not contribute to growth in the long run.
21
Consider the method in more detail. PSM uses a set of observed characteristics X to predict
the probability of an industry’s being chosen in the 10th PFYPs. The control group is then
constructed from the non-policy-selected industries to match the characteristics of the industries
selected in the 10th PFYPs. With a known score p(X) and the control group singled out by the PSM
method, the ATT of output after implementation of the 10th PFYPs is
𝑇𝑇𝐶𝐶𝐶𝐶 = 𝐶𝐶[𝑃𝑃𝑖𝑖1 − 𝑃𝑃𝑖𝑖0|𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 = 1] = 𝐶𝐶�𝐶𝐶�𝑃𝑃𝑖𝑖1 − 𝑃𝑃𝑖𝑖𝑖𝑖0|𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 = 1,𝑝𝑝(𝑇𝑇𝑖𝑖𝑖𝑖)�� = 𝐶𝐶�𝐶𝐶[𝑃𝑃𝑖𝑖𝑖𝑖1 |𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 = 1,𝑝𝑝(𝑇𝑇𝑖𝑖𝑖𝑖)]−
𝐶𝐶�𝑃𝑃𝑖𝑖𝑖𝑖0|𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 = 1,𝑝𝑝(𝑇𝑇𝑖𝑖𝑖𝑖)�|𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖 = 1�, (7)
where i indexes the industry and r indexes the province. When evaluating the contemporaneous
effect of the 10th PFYP, 𝑃𝑃𝑖𝑖𝑖𝑖1 is the output level of industry i in province r in 2005 (after treatment)
and 𝑃𝑃𝑖𝑖𝑖𝑖0 the output level in 2000 (before treatment). When evaluating the carry-over effect of the
10th PFYP, 𝑃𝑃𝑖𝑖𝑖𝑖1 is the output level of industry i in province r in 2010 (after treatment) and 𝑃𝑃𝑖𝑖𝑖𝑖0 the
output level in 2005 (before treatment). Since 𝑃𝑃𝑖𝑖𝑖𝑖0 is unobservable, we construct a counterfactual
to estimate ATT. To do this, industries which have not been favored during the 10th or the 11th
FYPs are matched with those in the treatment group, and are picked to replace the counterfactual
mean for those being treated. Since only variables that influence simultaneously the treatment
status and the outcome variable should be included (Smith and Todd, 2005), the characteristic
variables in Xir for industry i in region r used here are selected based on a balance test and those
finally included an industry dummy variable, a regional dummy variable, and a set of
pre-treatment variables like output share, tax, TFP, number of employees, number of firms and its
higher order forms, an industry dummy variable, and a regional dummy variable. 18 The
possibility of an industry in region r being favored in the 10th FYP is estimated with a logit model:
𝑝𝑝(𝑇𝑇𝑖𝑖𝑖𝑖) = 𝑃𝑃𝑃𝑃�𝑃𝑃𝑃𝑃𝑖𝑖,𝑠𝑠=0,𝑖𝑖 = 1|𝑇𝑇𝑖𝑖𝑖𝑖� = 𝑒𝑒𝑒𝑒𝑒𝑒(ℎ(𝑋𝑋𝑖𝑖𝑖𝑖))1+𝑒𝑒𝑒𝑒𝑒𝑒(ℎ(𝑋𝑋𝑖𝑖𝑖𝑖)) (8)
where ℎ(▪) is a polynomial function of the characteristic variable X.
(Insert Figure 7. around here)
Before considering the PSM-DID results, we look at some informal graphical evidence in
Figure 7 which shows the mean output of the treatment and control groups from 2000 to 2010
(left-hand axis) as well as the ratio of treatment to control group output (right-hand axis). The
steadily raising red and green dashed lines depict the trend of gross output of the treatment group
and that of the control group respectively, showing that the average scale of both groups grew
continuously since 2001, the first year of the 10th PFYPs. However the output ratio of the two
groups (the blue line) has an inverted U-shape, reaching its peak value in 2006, the first year after
18 Since sample size in this section is sharply reduced, we use dummy variables for the traditional division of China into four economic regions instead of a province dummy variable to ensure quality of matching. The eastern region includes Beijing, Tianjin, Hebei, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, Guangxi and Hainan; the central region includes Inner Mongolia, Shanxi, Henan, Anhui, Jiangxi, Hubei and Hunan; the western region includes Chongqing, Sichuan, Guizhou, Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia and Xinjiang; and the northeast region includes Liaoning, Heilongjiang, and Jilin.
22
the 10th PFYP. Thus, the treatment group had a higher average growth rate than the control group
during 2001-2005, when the former enjoyed favorable policy from the 10th PFYPs but after the
end of the plan the average growth rate of the treatment group was no longer faster than that of the
control group. Figure 7 gives a straightforward picture suggesting that the industrial policy does
not have a carryover effect. However, these are the unconditional effects likely to be biased
because the choice of the treatment might depend on or affect growth. So we now move to the
results of formal empirical analysis which accounts and corrects for the potential bias.
(Insert Table 17 around here)
(Insert Table 18 around here)
Table 17 and Table 18 show the estimation results for the standard two-period DID model as
specified in Equation (4), with and without sample screening through the PSM method. The key
parameter of interest is still the coefficient of the interaction term between policy and time.
Estimates in column (1) of Table 17 are the result of a standard DID model without sample
screening. It can be seen that the estimate of the interaction term is significant and positive,
indicating a beneficial influence of policy on industrial output during the 10th five-year period.
This is consistent with what we found in sections 5.1 and 5.2. We then use the PSM method to
screen samples and make sure the treatment group and the control group are similar at the baseline.
According to Abadie and Imbens (2006) and Baser (2006), there are several effective matching
techniques, the main ones being nearest-neighbor matching, radius matching, and kernel matching.
The estimates in column (2) are based on the matched samples using neighbor matching, where
the balancing is good for all covariates since the standardized bias is less than 5% and/or the
t-ratio is not significant for any covariates after matching.19 The estimated coefficient of the
interaction term remains significant and positive but the magnitude is larger than that of the
previous result. Recalling our previous discussion, this seems reasonable since the provincial
planner chooses his favorite industries according to the choice of the central planner and as a
result will reallocate resources to some industries which do not necessarily have local comparative
advantage. Therefore, the coefficient of the interaction term without sample screening will be
underestimated because of selection bias at the baseline. Moreover, if we take the logarithm of
industrial output as the dependent variable, the conclusion of an effective policy intervention
remains. This is shown in column (1) of Table 18. In brief, testing the effectiveness of the 10th
PFYP during its plan-period confirms our previous conclusion that policy stimulates the output of
preferred industries.
Having established the effectiveness of policy within the plan’s period of operation using the
19 The estimates in column (2) of Table 17 remain stable if other matching techniques, such as kernel matching or radius matching with caliper, are used.20 Note: ∀ i∈[1,419] and r∈[1,31], the value of the non-preferred group is the simple average of all non-preferred industries, the value of the preferred group is the simple average of all preferred industries, and the value of the total group is the simple average of all industries.
23
PSM-DID approach, we now turn to test the existence of a carry-over effect, i.e., whether there is
a persistent influence of preferential policy in the post-plan period. In column (3) of Table 17 we
show the DID estimates for the post-plan period of the interaction term without using PSM to
match samples. The relevant coefficient is positive and significant, indicating the existence of a
carry-over effect. However, if we employ the PSM method to screen samples, the coefficient of
the interaction term shown in column (4) is no longer significant and is much smaller than the
unmatched estimate. Judging by the dashed green and red lines in Figure 7, the estimate of the
simple DID method is biased due to the violation of the parallel-trend assumption. If the logarithm
of industrial output or output change is used as the dependent variable in the PSM-DID model as
reported in columns (2)-(4) of Table 18, the coefficient of the interaction term is no longer
significant and positive, irrespective of which matching technique is employed.
All in all, our results show that, unfortunately, on average the policy-promoted growth
observed during the currency of the relevant PFYP does not persist into the five-year period
beyond the end of the plan and so, policy does not have a long-lasting effect on industrial
development in China, unlike that found by Noland (1993) for Japan. Noland (1993) indicates that
industrial policies in Japan have had an impact on the trade pattern and have been forward-looking
in their policy interventions which have had persistent, long-lasting effects in Japan. Our
contrasting results for China are perhaps not surprising when referring to the TFP of the treatment
group and control group as illustrated in Figure 6. Since the preferred industries were neither those
with higher TFP level, nor those which had experienced a more rapid rate of TFP growth than that
of the control group during the period of the 10th PFYPs, it is natural, perhaps, to see that the
growth rate of the treatment group decreases in the period after the 10th PFYP.
To summarize the discussion in this sub-section, we have deepened our understanding of the
policy impact in the short and long terms and found that intervention by the government does
increase industry growth contemporaneously but that in the long term the development path
cannot easily be turned around. This suggests caution in the design of policy by provincial
planners.
5.4 Policy impact on industrial inputs
In the above three sub-sections, it has been confirmed that industrial policy embodied in the
FYP has a positive impact on contemporaneous output. Since industrial output could be viewed as
the ultimate outcome of policy intervention, the question arises as to whether industrial inputs,
such as labor and capital, are affected by the policy so as to finally result in increased output? In
this part we briefly examine the impact of policy embodied in the 10th and 11th PFYPs using the
regression specification defined in Equation (3) with various alternative dependent variables.
More specifically, we will test the effect of the 10th and 11th PFYPs on industrial employment
(measured by the logarithm of employee, lnL), the stock of capital (measured by the logarithm of
the real value of fixed assets, lnK), and the emergence of entrepreneurs (measured by the
logarithm of the number of firms, lnFirm). To keep the analysis simple, we use the same
24
sub-sample as in the estimation of column (5) in Table 13, which drops all the industries preferred
in both the 10th and 11th PFYPs as well as industries preferred in the 9th FYP.
(Insert Tables 19 near here)
(Insert Tables 20 near here)
The coefficient of the policy variable is again the key parameter of interest; the results are
reported in Table 19 and Table 20. From the first three columns in Table 19, it can be seen that
policy has a significant positive relationship with industrial employment and capital investment,
but does not show a statistically significant relationship with the emergence of entrepreneurs.
Further, the average firm size of preferred industries has increased significantly as a result of
preferential policy, whether measured by employees per firm (column (5) of Table 19) or by
capital per firm (column (4) of Table 19). Moreover, capital deepening (measured by the
logarithm of capital per employee, ln(K/L)) can be seen to be associated with preferred industries
for which capital per worker is increasing about 10% more than for the non-preferred industries.
If we take the regional location of the province into consideration when evaluating the policy
impact on industrial inputs which we do in Table 20, we find several differences between coastal
and inland provinces. For coastal provinces, policy did not have a significant association with
employment increases and capital deepening, although capital investment has been significantly
increased in the preferred industries (columns (1), (3) and (5) of Table 20), while among the
inland provinces, preferential policy was observed to have an obviously positive association with
employment, capital investment and capital deepening (columns (2), (4) and (6) of Table 20). The
reason for the differences in policy impact on industrial inputs between coastal and inland
provinces could lie in the differences in natural endowments and industry structure between them.
In the case of natural endowments, the inland provinces have more energy and resource industries,
which are capital intensive by nature. That is probably why we observe capital deepening in the
preferred industries in the economically lagging inland provinces compared to the more developed
coastal provinces in China.
In summary, the results presented above show that industrial policy stimulates industry
output mainly through increasing inputs like labor and capital rather than entrepreneurs, which
result in larger average firm size but a relatively stable number of firms. Industrial policy has a
positive association with capital deepening in the inland provinces, which might be associated
with differences in industrial structure between coastal and inland provinces rather than different
stages of industrial upgrading.
5.5 Further Robustness Checks
We have remarked several times above at the robustness of our results. Nevertheless, to
further check the robustness of the results, we carried out the following extra analysis.
(1) We re-ran all the regressions in Sections 4 and 5 without the top and bottom 5% of
25
observations as measured by industry gross output, to assess whether the reported estimates suffer
from extreme values.
(2) The simple average of firm-level TFP was used instead of the weighted average to compute
industry-level TFP for inclusion in the regressions in Table 11 and Table 12.
(3) The GMM method was used to account for possible heteroscedasticity for regressions (5) and
(6) in Table 15.
(4) For the PSM analysis, we also use radius matching with caliper to select matched samples.
In all cases the results are substantially unchanged.
6. Conclusions In China, the Five-Year Plan is one of the most significant sources of information to assess
whether and, if so, how regional governments take part in resource reallocation in an attempt to
influence the economic performance of particular industries in both the short term and the long
term. Based on the latest four rounds of FYPs and industry data at four-digit manufacturing sector
level, this paper has investigated the determinants and effectiveness of industrial policies
promulgated in successive provincial FYPs.
As to the determinants of choice embedded in the PFYPs, we found that the central
government’s preferences are a key determinant of the provincial governments’ choice of favored
industries when drafting their FYPs. We found this conclusion remarkably robust -- no matter
which other variables are used as controls, the impact of the central government’s FYP on the
possibility of selection by provincial governments dominates variables which measure industry
economic conditions at the beginning of the plan period. Moreover, the tendency for local
governments to follow the central government’s preference to select industries strengthens over
successive FYPs. Contrary to expectations, industry productivity is not a significant determinant
in selection of preferred industries, which means that the more efficient industries do not have a
larger chance of selection. We rationalized this by arguing that provincial governments are willing
to copy the central government’s preference at the cost of deviating from their own comparative
advantage, and/or aiming to support friends who happened to have lower productivity.
On the matter of the effectiveness of industry selection, we found that local industrial policy
does increase industry output through employment expansion and/or capital deepening during the
period for which the FYP is current. However, there is no statistical evidence that there is a
persistent, long-lasting impact on industrial development after the preferential policy ceases.
The findings of this paper shed light on the formation and effect of industrial policy. Though
the situation in China serves as the main motivation for the present paper, our analysis is general
enough to be referred to other countries characterized by extensive governmental intervention in
the allocation of resources at the industry level.
26
References:
Abadie, A. and G.W. Imbens, 2006. “Large sample properties of matching estimators for average
treatment effects.” Econometrica, Vol. 74, No. 1: 235-267.
Angrist, Joshua D., and Alan B. Krueger. 1999. Empirical strategies in labor economics. In
Handbook of Labor Economics, vol. 3A, ed. O. Ashenfelter and D. Card. Amsterdam:
North-Holland.
Alder, S., L. Shao, and F. Zilibotti, 2016. The effect of economic reform and industrial policy in a
panel of Chinese cities. Working paper. University of Zurich.
Bardhan, P., 1971. “On optimum subsidy to a learning industry: An aspect of the theory of
infant-industry protection”, International Economic Review, Vol.12: 54–70.
Baser, O., 2006. “Too much ado about propensity score models? Comparing methods of
propensity score matching.” Value in Health, Vol. 9, No. 6: 377-385.
Bastable, C. F., 1921. The Commerce of Nations, 10th edn. Macmillan and Co., London.
(Originally published in 1891.)
Belleflamme, Paul, Pierre Picard and Jacques F. Thisse. 2000. “An Economic Theory of Regional
Clusters.” Journal of Urban Economics 48: 158-184.
Chen, Z., S. Poncet, and R. Xiong, 2016. “Inter-industry relatedness and industrial-policy
efficiency: Evidence from China’s export processing zones”, Journal of Comparative Economics,
in press, http://dx.doi.org/10.1016/j.jce.2016.01.003.
Cheng, L. K. and Y. K. Kwan, 2000. “What are the determinants of the location of foreign direct
investment? The Chinese experience”, Journal of International Economics, Volume 51, Issue 2:
379-400.
Crozet, Matthieu, Thierry Mayer, and Jean L. Mucchielli. 2004. “How do firms agglomerate? A
study on FDI in France.” Regional Science and Urban Economics. 34:27-54.
Curtis, C. C., 2016. “Economic reforms and the evolution of China's total factor productivity”,
Review of Economic Dynamics, Vol. 21: 225-245.
Dethier, J., M. Hirn, and S. Straub, 2011. “Explaining enterprise performance in developing
countries with business climate survey data.” World Bank Research Observer, Vol 26, No.2:
258-309.
Fujita M., P. Krugman and A. Venables, 1999, The Spatial Economy, The MIT Press, Cambridge.
Hamilton, A., 1791. “Report on Manufactures, December 5”, in The Papers of Alexander
Hamilton. Edited by Harold C. Syrett et al. Columbia University Press, New York and London:
252-256.
Harrison, A. E., 1994. “An empirical test of the infant industry argument: Comment”. American
Economic Review, Vol. 84, No. 4: 1090–1095.
Harrison, A., and A. Rodríguez-Clare, 2010. “Trade, foreign investment, and industrial policy for
developing countries”, Handbook of Development Economics, Vol. 5, Elsevier, Amsterdam.
Hausmann, R., J. Hwang and D. Rodrik, 2007. “What you export matters”, Journal of Economic
27
Growth, Vol. 12, No. 1:1-25.
Head,Keith,John Ries and Deborah Swenson. 1995. “agglomeration benefits and locatioin
choice: Evidence from Japanese manufacturing investments in the United States.” Journal of
International Economics, 38:223-247.
Hu, A., 2011, “The democratization of public policy making with Chinese characteristic:
Formulation of China’s 12th Five-Year Plan”, Journal of Tsinghua University (Philosophy and
Social Sciences), (in Chinese), Vol. 26, No. 2:43-50.
Hua, X., X. Liu and M. Wang, 2016. “Industrial policies and innovation: Evidence from the
Chinese automotive and electronics industries”, Working Paper, Nottingham University Business
School, China.
Jefferson, G. H., T. G. Rawski, L. Wang, and Y. Zheng, 2000. “Ownership, Productivity change,
and financial performance in Chinese industry”, Journal of Comparative Economics, Vol.
28:.786-813.
Krueger, A. O., and B. Tuncer, 1982. “An empirical test of the infant industry argument”,
American Economic Review, Vol. 72, No. 5: 1142–1152.
Krugman P. 1991, “Increasing returns and economic geography”, Journal of Political Economy,
Vol. 99, No. 3: 483-499.
Laurenceson, J., and J. C. H. Chai, 2000. “The economic performance of China’s state-owned
industrial enterprises”. Journal of Contemporary China, Vol. 9: 21-39.
Lee, E., M. Walker and C. Zeng, 2014. “Do Chinese government subsidies affect firm value?”,
Accounting Organization and Society. Vol. 39: 149-169.
Leuven E., and B. Sianesi. 2003, “PSMATCH2: Stata module to perform full Mahalanobis and
propensity score matching, common support graphing, and covariate imbalance testing”.
Available at: http://ideas.repec.org/c/boc/bocode/s432001.html.
Lin, J., 2012, New Structural Economics: A Framework for Rethinking Development and Policy,
Beijing University Press, Beijing. (in Chinese)
List, F. 1856. The National System of Political Economy, JB Lippincott and Co., Philadelphia.
Lu, J. and Z. Tao, 2009, “Trends and determinants of China’s industrial agglomeration”, Journal
of Urban Economics, Vol. 65: 167-180.
Nie, H., T Jiang and R. Yang, 2012, “A review and reflection on the use and abuse of the Chinese
industrial enterprises database”. World Economy No. 5: 142-158. (in Chinese),.
Marshall, A. 1920, Principles of Economics, MacMillan, London.
Melitz, M. 2005, “When and how should infant industries be protected?”, Journal of International
Economics, Vol. 66: 177–196.
Mill, J. 1848, Principles of Political Economy, Longman, London.
Mora, R., and I. Reggio, 2012. “Treatment effect identification using alternative parallel
assumptions.” Working aper 12-33, Economic Series (48), Universidad Carlos III de Madrid,
Noland, M. 1993, “The impact of industrial policy on Japan’s trade specialization”, Review of
28
Economics and Statistics, Vol. 75, No. 2: 241-248.
Olley, G. S. and A. Pakes, 1996. “The dynamics of productivity in the telecommunications
equipment industry, Econometrica, Vol. 64: 1263-1297.
Pack, H. 2000, “Industrial policy: Growth elixir or poison?”, The World Bank Research Observer,
Vol. 15, No. 1: 47-67.
Pack, H. and K. Saggi, 2006, “Is there a case for industrial policy? A critical survey”, The World
Bank Research Observer, Vol. 21, No. 2: 267-297.
Porter, M. 1990, The Competitive Advantage of Nations, Free Press, New York.
Redding, S. 1999, “Dynamic comparative advantage and the welfare effects of trade”, Oxford
Economic Papers, Vol. 51, No. 1: 15–39.
Rodrik, D. 2004, “Industrial policy for the twenty-first century”, Harvard University.
Rodrik, D. 2006, “What’s special about China’s exports?” China and World Economy, Vol. 14: 1–
19.
Rodrik, D.2009, “Industrial policy: Don’t ask why, ask how”, Middle East Development Journal,
Vol. 1, No. 1: 1-29.
Rosenbaum, P. R. and D. B. Rubin, 1983, “The central role of the propensity core in observational
studies for causal effects”, Biometrika, Vol. 70, No. 1: 41-55.
Smith, J. A., and P. E. Todd, 2005, “Dose matching overcome LaLonde’s critique of
nonexperimental estimators?” Journal of Econometrics, Vol. 125: 305-353.
Storesletten, K., and F. Zilibotti, 2014, “China’s great convergence and beyond”, Annual Review
of Economics, Vol. 6:333-362.
Song, L. and X. Wang, 2013, “Industry policy, resource allocation and productivity”, Management
World Vol. 12: 63-77. (In Chinese)
Wang, J. 2013, “The economic impact of Special Economic Zones: Evidence from Chinese
municipalities”, Journal of Development Economics, Vol. 101:133-147.
Wei, S.J., 1995, “The open door policy and China's rapid growth: evidence from city-level data”.
In: Ito, T. and A. O. Krueger (Eds.), Growth Theories in Light of the East Asian Experience.
University of Chicago Press, Chicago: 73-98.
Wooldridge, J. M. 1995. “Score diagnostics for linear models estimated by two stage least
squares”. In G. S. Maddala, P. C. B. Phillips, and T. N. Srinivasan (eds) Advances in
Econometrics and Quantitative Economics: Essays in Honor of Professor C. R. Rao, Blackwell,
Oxford: 66-87.
Wooldridge, J. M. 2006, Introductory Econometrics: A Modern Approach, 3nd edition,
South-Western (Cengage Learning).
Wu Y., and X., Zhu. 2015, “Why dispersion before its time in China: Industry policy and
economic geography”, World Economy, Vol. 2: 140-166. (in Chinese)
Zhang F., 2014, Industry Drift: Geographical Transfer of the World Manufacturing Industries and
the Central Market, Beijing University Press, Beijing. (in Chinese)
29
Zhou L., Y. Chen and H. Li 2005, "Relative performance evaluation and the turnover of provincial
leaders in China", Economics Letters, Vol. 88: 421-425.
30
Table 1. Period covered and year of issue of China’s thirteen National Five-Year Plans
Plan number
Period covered
Month of approval Plan number
Period covered
Month of issue
1st Plan 1953–1957 1955.7 8th Plan 1991–1995 1991.4
2nd Plan 1958–1962 1958.8 9th Plan 1996–2000 1996.3
3rd Plan 1966–1970 Not formally issued 10th Plan 2001–2005 2001.3
4th Plan 1971–1975 1970, revised in 1973 11th Plan 2006–2010 2006.3
5th Plan 1976–1980 1977.12 12th Plan 2011–2015 2011.3
6th Plan 1981–1985 1982.12 13th Plan 2016-2020 2016.3
7th Plan 1986–1990 1986.3
Table 2. Number of policy-preferred 4-digit industries in the 9th to 12th FYPs
Preferred industries
PFYPs NFYPs 9th 10th 11th 12th Mean 9th 10th 11th 12th Mean
Number 2913 3208 2472 2685 2820 115 180 140 178 153 % of total industries 22.43 24.70 19.03 20.67 21.71 27.44 42.96 33.41 42.48 36.52
Note; the number of industries in the PFYP columns are summed across all provinces
31
Table 3. Top 30 policy-preferred 4-digit manufacturing industries in the 9th to 12th PFYPs
Industry code
Industry name PFYP 9th 10th 11th 12th
2760 Biological, biochemical products manufacturing 28 31 29 30 2710 Chemical pharmaceutical raw material manufacturing 30 31 29 28 2720 Chemical pharmaceutical manufacturing 30 31 29 28 3721 Automobile manufacturing 27 18 24 27 2730 Chinese herbs slices and decoction processing 27 31 30 26 2750 Veterinary drug manufacturing 25 31 29 26 2770 Hygiene materials and medical supplies manufacturing 25 31 29 26 2665 Photographic chemicals manufacturing 25 20 29 26 1352 Meat products and by-products processing 11 20 20 26 4061 PCB and electronic components manufacturing 27 28 29 25 3725 Auto parts and accessories manufacturing 28 18 26 25 1494 Food and feed additives manufacturing 26 28 22 25 2631 Chemical pesticide manufacturing 24 25 18 25 2671 Soap and synthetic detergent manufacturing 25 26 17 25 1320 Feed processing 24 26 17 25 2619 Other chemical manufacturing 24 26 17 25 2672 Cosmetics manufacturing 25 25 17 25 2641 Paint manufacturing 24 25 17 25 2645 Sealing packing and similar products manufacturing 24 25 17 25 2644 Dye manufacturing 23 25 17 25 2661 Chemical reagent and catalysts manufacturing 24 22 17 25 2662 Special chemical products manufacturing 23 22 17 25 3220 Steel making 22 10 23 24 2511 Crude oil processing and petroleum products 18 16 22 24 2674 Flavor and fragrance manufacturing 24 19 18 24 2642 Ink and similar products manufacturing 23 25 17 24 2643 Pigment manufacturing 23 25 17 24 2679 Other daily chemical products manufacturing 22 21 17 24 1440 Liquid milk and dairy products manufacturing 10 20 20 22 1810 Textile and garment manufacturing 18 20 18 22 Note: (1) numbers in the last four columns represent the number of provinces which selected that particular industry in that particular PFYP; (2) industries are sorted according to times selected in the 12th PFYPs; (3) industry 2619 is a combination of Other basic chemical raw materials manufacturing (2619), Other specialized chemical products manufacturing (2669) and Manufacture of special chemicals for environmental pollution treatment (2666); (4) industry 4061 is a combination of Printed circuit board manufacturing (4061) and Electronic components manufacturing (4062); (5) The ranking of preferred industry is generally highly consistent through time except for industries within the two-digit Chemical raw materials & chemical products manufacturing industry (with two-digit code 26). The fluctuation in the number of provinces which selected these industries in the 11th PFYPs is noticeable and could be associated with a sudden reduction of 25% in numbers of preferred industries in this two-digit industry in the 11th NFYP, compared to the other NFYPs. The selection of PFYP are highly correlated with that of the NFYP; we provide evidence for this relationship between NFYPs and PFYPs in Table 6 and in section 4.
32
Table 4. Spearman rank correlation coefficient for industry rankings in Table 3
9th PFYP 10th PFYP 11th PFYP 12th PFYP
9th PFYP 1 10th PFYP 0.7455 1
11th PFYP 0.7084 0.7937 1 12th PFYP 0.7242 0.6912 0.8308 1
Table 5. Number of policy-preferred industries in the 9th to 12th PFYPs
Policy-preferred industries 9th 10th 11th 12th Average Maximum 182 201 138 155 169 Minimum 19 16 28 20 21 Mean: nationwide 94 103 80 91 92 Mean: coastal region 105 134 88 108 109 Mean: central region 106 105 90 75 94 Mean: western region 70 66 61 68 66 Note: the coastal region consists of Liaoning, Beijing, Tianjin, Hebei, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, Guangxi and Hainan; the central region consists of Heilongjiang, Jilin, Inner Mongolia, Shanxi, Henan, Anhui, Jiangxi, Hubei and Hunan; the western region consists of Chongqing, Sichuan, Guizhou, Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia and Xinjiang. Table 6. Number of provinces selecting industries compared to the 9th to 12th NFYPs
9th 10th 11th 12th 9th v. 12th(t-statistic) For industries preferred in NFYPs 11.5 12.3 13 16.5 4.05*** For industries not preferred in NFYPs 4.9 4.1 2.3 3.9 -2.45*** For all the manufacturing industries 6.8 7.6 5.8 6.2 -1.18 Table 7 Construction of CTYPEin based on the two preceding NFYPs
Industry i Status in the (n-1)th NFYP(2000-2005)
Status in the nth NFYP(2006-2010)
CTYPEin =1 if Non-preferred Non-preferred CTYPEin =2 if Preferred Non-preferred CTYPEin =3 if Non-preferred Preferred CTYPEin =4 if Preferred Preferred
33
Table 8. Determinants of industry selection in PFYPs
(1) (2) (3) (4) (5)
CTYPEin: =2 0.315** 0.313** 0.311** 0.306** 0.305**
(0.151) (0.151) (0.151) (0.152) (0.151)
=3 0.491*** 0.487*** 0.484*** 0.488*** 0.486***
(0.072) (0.072) (0.072) (0.072) (0.072)
=4 1.352*** 1.339*** 1.336*** 1.332*** 1.330***
(0.092) (0.092) (0.092) (0.093) (0.093)
EMP 0.023*** 0.018*** 0.011** 0.009*
(0.007) (0.006) (0.005) (0.005) TAXFEE 0.215*** 0.140***
(0.065) (0.050) OPSHARE 0.062*** 0.054***
(0.019) (0.019) Perd_dummy YES YES YES YES YES
Prov_dummy YES YES YES YES YES
Ind_dummy YES YES YES YES YES
_cons -0.592*** -0.592*** -0.591*** -0.625*** -0.620***
(0.152) (0.152) (0.153) (0.154) (0.154)
N 22113 22113 22113 22113 22113
LR chi2 3420.30 (0.000)
3529.64 (0.000)
3450.24 (0.000)
3522.45 (0.000)
3534.63 (0.000)
Pseudo R2 0.2653 0.2667 0.2682 0.2698 0.2703 Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01. (3) cluster-robust standard errors (cluster at 4-digit industry level) are used. Table 9 Predicted probabilities and marginal effects of the main variables in Table 8
Margin Std. Err. Z P>z 95% Conf. Interval Predicted probabilities: CTYPEin=1 0.0985 0.0102 9.63 0.000 0.0785 0.1186 =2 0.1624 0.0340 4.78 0.000 0.0958 0.2290 =3 0.2108 0.0127 16.55 0.000 0.1858 0.2358 =4 0.5161 0.0246 20.97 0.000 0.4678 0.5644 Marginal effect: CTYPEin=2 0.0639 0.0357 1.79 0.073 -0.0060 0.1337 =3 0.1122 0.0161 7.00 0.000 0.0808 0.1437 =4 0.4176 0.0283 14.77 0.000 0.3622 0.4730
EMP 0.0025 0.0014 1.76 0.079 -0.0003 0.0053 OPSHARE 0.0154 0.0054 2.83 0.005 0.0047 0.0261 TAXFEE 0.0400 0.0145 2.77 0.006 0.0117 0.0683
34
Table 10. The role of regions in the determinants of industry selection in PFYPs
Dependent. Variable: 𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖
(1)Probit (2) Probit (3) Probit (4) Probit (5) Probit
CTYPEin: =2 0.508*** 0.510*** 0.510*** 0.508*** 0.508***
(0.176) (0.175) (0.175) (0.174) (0.174)
=3 0.550*** 0.541*** 0.539*** 0.541*** 0.540***
(0.067) (0.067) (0.068) (0.067) (0.068)
=4 1.404*** 1.383*** 1.380*** 1.378*** 1.377***
(0.092) (0.093) (0.093) (0.093) (0.093)
inland× =1 0.455*** 0.453*** 0.458*** 0.432*** 0.438***
CTYPEin: (0.116) (0.115) (0.115) (0.115) (0.115)
=2 0.0892 0.0784 0.0795 0.0473 0.0520
(0.196) (0.196) (0.195) (0.195) (0.195)
=3 0.350*** 0.356*** 0.359*** 0.336*** 0.341***
(0.109) (0.108) (0.109) (0.108) (0.108)
=4 0.361*** 0.373*** 0.378*** 0.347*** 0.353***
(0.112) (0.112) (0.112) (0.111) (0.111)
EMP 0.0226*** 0.0174*** 0.0103** 0.00862*
(0.007) (0.006) (0.005) (0.005) TAXFEE 0.217*** 0.141***
(0.065) (0.050) OPSHARE 0.0618*** 0.0541***
(0.019) (0.019) Perd_dummy YES YES YES YES YES
Prov_dummy YES YES YES YES YES
Ind_dummy YES YES YES YES YES _cons -0.637*** -0.633*** -0.632*** -0.668*** -0.663*** (0.153) (0.153) (0.154) (0.155) (0.155) N 22113 22113 22113 22113 22113
LR chi2 3746.28 (0.000)
3856.54 (0.000)
3720.10 (0.000)
3813.83 (0.000)
3833.39 (0.000)
Pseudo R2 0.2657 0.2670 0.2685 0.2701 0.2706 Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01. (3) cluster-robust standard errors (cluster at 4-digit industry level) are used.
35
Table 11. Industry selection and TFP in the 10th and 11th PFYPs
Dependent. Variable: 𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖
(1) Probit (2) Probit (3) Probit (4) Probit (5) Probit
CTYPEin: =2 0.351* 0.345* 0.344* 0.339* 0.339* (0.184) (0.183) (0.182) (0.181) (0.181)
=3 0.558*** 0.555*** 0.553*** 0.553*** 0.552*** (0.088) (0.089) (0.089) (0.089) (0.089)
=4 1.343*** 1.331*** 1.330*** 1.323*** 1.323*** (0.109) (0.110) (0.110) (0.111) (0.111) TFP -0.063*** -0.056*** -0.057*** -0.048** -0.049**
(0.020) (0.020) (0.020) (0.019) (0.019) EMP 0.020*** 0.014** 0.007 0.005 (0.00651) (0.00620) (0.006) (0.006) TAXFEE 0.331** 0.185* (0.138) (0.113) OPSHARE 0.056*** 0.050***
(0.017) (0.016) Perd_dummy YES YES YES YES YES
Ind_dymmy YES YES YES YES YES Prov_dummy YES YES YES YES YES _cons -0.396** -0.403** -0.402** -0.431** -0.428** (0.180) (0.180) (0.180) (0.181) (0.181)
N 11191 11191 11191 11191 11191
LR chi2 2962.21 (0.000)
3183.69 (0.000)
3096.24 (0.000)
3257.97 (0.000)
3235.03 (0.0000)
Pseudo R2 0.3095 0.3103 0.3115 0.3134 0.3137 Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01. (3) cluster-robust standard errors (cluster at 4-digit industry level) are used. (3) TFP data are not available after 2008, so that the 12th PFYP is excluded from the sample used in Table 11.
36
Table 12. Industry selection and TFP in the 10th and 11th PFYPs, contd.
Dependent. Variable: 𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖
(6) Probit (7) Probit (8) Probit (9) Probit (10) Probit
CTYPEin: =2 0.314* 0.312* 0.309* 0.304* 0.303* (0.188) (0.188) (0.187) (0.184) (0.184)
=3 0.488*** 0.489*** 0.486*** 0.486*** 0.486*** (0.089) (0.089) (0.089) (0.089) (0.089)
=4 1.296*** 1.298*** 1.291*** 1.278*** 1.275*** (0.110) (0.110) (0.110) (0.111) (0.112) H-TFP80 -0.089**
(0.045) H-TFP90 -0.062 -0.064 -0.049 -0.047 (0.054) (0.054) (0.054) (0.054) TAXFEE 0.465*** 0.248* 0.227* (0.170) (0.128) (0.125) OPSHARE 0.057*** 0.054***
(0.0180) (0.0177) EMP 0.007
(0.00638) Perd_dummy YES YES YES YES YES
Ind_dymmy YES YES YES YES YES
Prov_dummy YES YES YES YES YES
_cons -0.375** -0.381** -0.380** -0.401** -0.398**
(0.178) (0.179) (0.179) (0.181) (0.181)
N 12340 12340 12340 12340 12340
LR chi2 2831.64 (0.000)
2835.49 (0.000)
2842.75 (0.000)
2987.71 (0.000)
3016.30 (0.0000)
Pseudo R2 0.3060 0.3057 0.3078 0.3106 0.3107 Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01. (3) cluster-robust standard errors (cluster at 4-digit industry level) are used. (3) TFP data are not available after 2008, so that the 12th PFYP is excluded from the sample used in Table 12.
37
Table 13. The output effects of industry selection in the 10th and 11th PFYPs
(1) DID lnY
(2) DID lnY
(3) DID lnY
(4) DID lnY
(5) DID lnY
𝑃𝑃𝑃𝑃𝑖𝑖𝑖𝑖𝑖𝑖10&11 0.110*** 0.0795** 0.0953** 0.134*** 0.145*** (0.0376) (0.0340) (0.0383) (0.0373) (0.0438) Period: 2005 0.794*** 0.796*** 0.792*** 0.811*** 0.828*** (0.0283) (0.0262) (0.0266) (0.0286) (0.0317) Period: 2010 1.239*** 1.204*** 1.288*** 1.243*** 1.263*** (0.0464) (0.0429) (0.0410) (0.0476) (0.0496) Treat I 0.316*** 0.224*** 0.274*** (0.0750) (0.0624) (0.0688) Treat II 0.709*** 0.383*** 0.660*** 0.282*** (0.138) (0.0984) (0.131) (0.0739) Treat II-I 0.0639 (0.0832) Treat II-II 0.274** (0.110) Ind_dymmy YES YES YES YES YES
Prov_dummy YES YES YES YES YES _cons -1.628*** -1.804*** -1.422*** -1.616*** -1.698*** (0.345) (0.298) (0.308) (0.385) (0.413) N 22113 20959 21052 18764 14608 adj. R2 0.340 0.342 0.380 0.341 0.346 F-stat 84.81 95.61 88.79 81.80 60.57 (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01. (3) cluster-robust standard errors (cluster at 4-digit industry level) are used.
38
Table 14. The output effects of industry selection in the 11th PFYP
(1)DID (2)DID (3)DID (4)DID (5)DID (6)DID lnY lnY lnY ln∆Y ln∆Y ln∆Y 𝑃𝑃𝑃𝑃 ∗ 𝐶𝐶𝑇𝑇 0.175*** 0.172*** 0.172*** 0.200*** 0.192*** 0.192*** (0.0515) (0.0474) (0.0474) (0.0637) (0.0630) (0.0630) 𝑃𝑃𝑃𝑃 0.488*** 0.330*** 0.330*** 0.414*** 0.271*** 0.271*** (0.106) (0.0822) (0.0822) (0.111) (0.0915) (0.0915) 𝐶𝐶𝑇𝑇 0.401*** 0.382*** 0.382*** 0.409*** 0.397*** 0.397*** (0.0387) (0.0363) (0.0363) (0.0422) (0.0403) (0.0403) FIRM 0.0189*** 0.0189*** 0.0169*** 0.0169*** (0.00144) (0.00144) (0.00128) (0.00128) FIRMsq -0.0941*** -0.0941*** -0.0830*** -0.0830*** (0.0158) (0.0158) (0.0138) (0.0138) D_inland -0.653*** -0.395*** (0.129) (0.131) Ind_dummy YES YES YES YES YES YES Prov_dummy YES YES YES YES YES YES _cons -0.800** -1.246*** -1.246*** -1.344*** -1.764*** -1.764*** (0.362) (0.232) (0.232) (0.338) (0.227) (0.227) N 16117 14764 14764 14662 13309 13309 adj. R2 0.306 0.463 0.463 0.281 0.413 0.413 F-stat 62.12
(0.0000) 52.96
(0.0000) 52.96
(0.0000) 49.42
(0.0000) 58.81
(0.0000) 58.81
(0.0000) Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01; (3) r cluster-robust standard errors (cluster at 4-digit industry level) are used.
39
Table 15. The output effects of industry selection in the 11th PFYP: Controlling for
multi-period selection
(1) DID (2) DID (3) DID (4) DID (5) 2SLS (6) 2SLS lnY lnY ln∆Y ln∆Y lnY lnY 𝑃𝑃𝑃𝑃 ∗ 𝐶𝐶𝑇𝑇 0.202*** 0.213*** 0.243*** 0.245*** (0.069) (0.067) (0.083) (0.085) 𝑃𝑃𝑃𝑃 0.347*** 0.223*** 0.282** 0.172* 0.222*** 0.243*** (0.110) (0.0831) (0.114) (0.095) (0.081) (0.080) 𝐶𝐶𝑇𝑇 0.428*** 0.423*** 0.442*** 0.444*** (0.0418) (0.0388) (0.045) (0.043) FIRM 0.020*** 0.0175*** 0.0031*** (0.001) (0.001) (0.0002) FIRMsq -0.095*** -0.084*** -0.015*** (0.015) (0.012) (0.002) lnYn-1 0.856*** 0.793*** (0.008) (0.010) Ind_dummy YES YES YES YES YES YES Prov_dummy YES YES YES YES YES YES _cons -0.755* -1.302*** -1.364*** -1.877*** 0.268*** 0.122* (0.440) (0.270) (0.412) (0.274) (0.070) (0.071) N 11319 10246 10368 9295 6510 6510 adj. R2 0.308 0.458 0.285 0.411 0.787 0.793 F-stat 46.70
(0.0000) 48.61
(0.0000) 38.34
(0.0000) 56.44
(0.0000)
Wald chi2
22896 (0.0000)
22896 (0.0000)
IV variables NA NA NA NA Ctype2i Ctype2i 1st stage robust F statistics
686.064 (0.0000)
681.581 (0.0000)
1st stage Shea’s adj. Partial R-sq.
0.1115 0.1108
DHW (test of endogeneity)
4.23982 (0.0395)
5.05886 (0.0245)
Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01; (3) cluster-robust standard errors (cluster at 4-digit industry level) are used. Table 16. Definition of Ctype2 based on the 9th to 11th PFYPs and NFYPs
Industry i Status in the 9th NFYP
Status in the 10th NFYP
Status in the 11th NFYP
Ctype2i=1 if Non-preferred Non-preferred Non-preferred Ctype2i=2 if Preferred Non-preferred Non-preferred Ctype2i=3 if Non-preferred Preferred Non-preferred Ctype2i=4 if Preferred Preferred Non-preferred Ctype2i=5 if Non-preferred Non-preferred Preferred Ctype2i=6 if Preferred Non-preferred Preferred Ctype2i=7 if Non-Preferred Preferred Preferred Ctype2i=8 if Preferred Preferred Preferred
40
Table 17. PSM-DID estimates of the short-term and long-tern output effects of industry
selection
(1) DID (2) PSM-DID (3) DID (4) PSM-DID (Neighbor) (Neighbor) Y Y Y Y Short-term (2000-2005) Long-term (2006-2010) 𝑃𝑃𝑃𝑃 ∗ 𝐶𝐶𝑇𝑇 0.734** 1.491** 1.556** 1.061 (0.282) (0.660) (0.647) (0.728) 𝑃𝑃𝑃𝑃 0.263 0.212 0.652 0.611 (0.217) (0.253) (0.402) (0.494) 𝐶𝐶𝑇𝑇 0.416*** 1.882*** 1.493*** 2.997*** (0.101) (0.343) (0.203) (0.454) Region dummy Yes Yes Yes Yes Industry dummy Yes Yes Yes Yes _cons 1.587*** 1.230** 2.922*** 3.364*** (0.355) (0.535) (0.620) (1.002) N 14384 3134 21034 5682 adj. R2 0.040 0.040 0.042 0.036 F-stat 12.93 17.85 25.69 14.14 (0.0000) (0.0000) (0.0000) (0.0000) Note: (1)* for p < 0.10, ** for p < 0.05, *** for p < 0.01; (2) the propensity score is estimated with the user-written program psmatch2 (Leuven and Sianesi, 2003) in stata13. (3) the balance testing is conducted with the command pstest. (4) due to sample screening, the industry dummy in Table 17 and Table 18 is set according to the industrial category on page 259 of Zhang (2014).
Table 18. RPSM-DID estimates of the short-term and long-term output effects of industry
selection: Alternative matching assumptions
(1) PSM-DID (Neighbor)
(2) PSM-DID (Neighbor)
(3) PSM-DID (Kernel)
(4) PSM-DID (Neighbor)
lnY lnY ln∆Y ln∆Y Valid Period (2000-2005) Ex post (2006-2010) 𝑃𝑃𝑃𝑃 ∗ 𝐶𝐶𝑇𝑇 0.0675* -0.0108 -0.0940 0.0347 (0.037) (0.065) (0.0860) (0.0950) 𝑃𝑃𝑃𝑃 0.0844 -0.0475 0.0996 -0.0516 (0.137) (0.116) (0.112) (0.131) 𝐶𝐶𝑇𝑇 0.804*** 0.519*** 0.612*** 0.487*** (0.047) (0.072) (0.0730) (0.0940) Region dummy Yes Yes Yes Yes Industry dummy Yes Yes Yes Yes _cons -0.753*** 0.103 -0.508** -0.415** (0.232) (0.199) (0.192) (0.192) N 3134 4977 9924 4439 adj. R2 0.174 0.108 0.097 0.086 F-stat 121.93 37.63 31.12 38.38 (0.0000) (0.0000) (0.0000) (0.0000) Note: (1)* for p < 0.10, ** for p < 0.05, *** for p < 0.01; (2) the propensity score is estimated the with user-written program psmatch2 (Leuven and Sianesi, 2003) in stata13. (3) the balance testing is conducted with the command pstest. (4) due to sample screening, the industry dummy in Table 17 and Table 18 is set according to the industrial category on page 259 of Zhang (2014).
41
Table 19. The channels of industrial policy in the 10th and 11th PFYPs
(1) (2) (3) (4) (5) (6) lnL lnK lnFirm lnK/Firm lnL/Firm lnK/L PPirn 0.129*** 0.229*** 0.0228 0.102*** 0.0415* 0.104*** (0.0402) (0.0490) (0.0237) (0.0331) (0.0213) (0.0259) Period: 2005 0.00247 0.179*** 0.229*** -0.0365 -0.220*** 0.177*** (0.0289) (0.0346) (0.0252) (0.0253) (0.0192) (0.0170) Period: 2010 -0.287*** 0.223*** 0.566*** 0.244*** -0.344*** 0.507*** (0.0458) (0.0544) (0.0317) (0.0289) (0.0223) (0.0217) Treat II-I 0.0699 0.0385 0.0504 -0.000786 0.0302 -0.0372 (0.0748) (0.0863) (0.0620) (0.0455) (0.0341) (0.0290) Treat II-II 0.289*** 0.316*** 0.0994 0.185*** 0.183*** 0.0263 (0.0952) (0.117) (0.0786) (0.0665) (0.0483) (0.0377) Ind_dummy YES YES YES YES YES YES Prov_dummy YES YES YES YES YES YES _cons 7.196*** 11.51*** 2.589*** 8.988*** 4.638*** 4.319*** (0.320) (0.410) (0.325) (0.124) (0.102) (0.117) N 14608 14595 13018 13017 13018 14595 adj. R2 0.267 0.244 0.318 0.176 0.155 0.234 F-stat 53.75 35.25 23.96 16.45 23.33 31.59 (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01; (3) cluster-robust standard errors (cluster at 4-digit industry level) are used. Table 20. The channels of industrial policy in the 10th and 11th PFYPs: Regional
differences
(1)coastal (2)inland (3) coastal (4) inland (5) coastal (6)inland lnL lnL lnK lnK lnK/L lnK/L PPirn 0.0764 0.204*** 0.118** 0.322*** 0.0412 0.126*** (0.0464) (0.0613) (0.0581) (0.0730) (0.0337) (0.0368) Period: 2005 0.144*** -0.124*** 0.274*** 0.0989** 0.130*** 0.224*** (0.0379) (0.0344) (0.0451) (0.0412) (0.0199) (0.0225) Period: 2010 0.00470 -0.479*** 0.350*** 0.148** 0.344*** 0.622*** (0.0542) (0.0531) (0.0646) (0.0622) (0.0238) (0.0263) Treat II-I 0.0656 0.0461 0.0171 0.0509 -0.0485 -0.00779 (0.0949) (0.0872) (0.111) (0.102) (0.0388) (0.0381) Treat II-II 0.301** 0.231** 0.321* 0.262** 0.0211 0.0281 (0.149) (0.108) (0.173) (0.132) (0.0609) (0.0438) Ind_dummy YES YES YES YES YES YES Prov_dummy YES YES YES YES YES YES _cons 6.766*** 7.504*** 11.21*** 11.51*** 4.449*** 4.007*** (0.294) (0.376) (0.369) (0.472) (0.132) (0.129) N 6207 8401 6206 8389 6206 8389 adj. R2 0.273 0.228 0.239 0.212 0.255 0.237 F-stat 34.77 32.51 23.64 19.59 31.82 34.57 (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) (0.0000) Note: (1) for variables, standard errors in parentheses and for statistics, p value in parentheses; (2) * p < 0.10, ** p < 0.05, *** p < 0.01; (3) cluster-robust standard errors (cluster at 4-digit industry level) are used.
42
Figure 1. Firm scale and policy selection in the 10th - 12th PFYPs20
Figure 2. Number of employees per industry and policy selection in the 10th - 12th PFYPs
20 Note: ∀ i∈[1,419] and r∈[1,31], the value of the non-preferred group is the simple average of all non-preferred industries, the value of the preferred group is the simple average of all preferred industries, and the value of the total group is the simple average of all industries.
43
Figure 3. Gross output and policy selection in the 10th - 12th PFYPs
Figure 4. Average fixed-assets and policy selection in the 10th - 12th PFYPs
44
Figure 5. Value added and policy selection in the 10th - 11th PFYPs21
Figure 6. Industry TFP and policy selection in the 10th - 11th PFYPs22
21 Value-added by industry and province is not available for 2010. 22 Firm level FTP for 2010 is not available.
45
Figure 7. Trends of Gross Output
46
Appendix: Summary of main regression variables Variable Explanation Mean S.E. EMP number of employees in industry i, province r, 10000
persons 0.631 2.784
OPSHARE output of industry i, province r divided by total output of province r, %
0.239 1.111
TAXFEE total tax and extra charges on sales of products plus income tax of industry i, province r, billion yuan
0.051 0.411
PFYP binary variable, identifying whether industry i is in or out (1 for in and 0 for out) of province r’s preferred-industry list in the nth PFYP
0.190 0.393
Ln∆Y logarithm of output differences over five years -0.346 2.165 LnY logarithm of output -0.652 2.097 FIRM number of firms 42.59 131.8 FIRMsq square of FIRM/100 1.199 26.86 CTYPEin categorical variable, see Table 7 --- --- Ctype2i categorical variable, see Table 16 --- --- Perd_dummy period dummy variables --- --- Prov_dummy province dummy variables --- --- Ind_dummy industry dummy variables --- --- Region dummy region dummy variables --- --- TFP total factor productivity for industry; data for 2010 is
missing --- ---
H-TFP90irn binary variable which equals 1 for industry i in province r if its TFP ranks in the top 0.10 quantile of that province and 0 for the others; data for 2010 is missing
--- ---
H-TFP80irn binary variable which equals 1 for industry i in province r if its TFP ranks in the top 0.20 quantile of that province and 0 for the others; data for 2010 is missing
--- ---
47
Editor, UWA Economics Discussion Papers: Sam Hak Kan Tang University of Western Australia 35 Sterling Hwy Crawley WA 6009 Australia Email: [email protected] The Economics Discussion Papers are available at: 1980 – 2002: http://ecompapers.biz.uwa.edu.au/paper/PDF%20of%20Discussion%20Papers/ Since 2001: http://ideas.repec.org/s/uwa/wpaper1.html Since 2004: http://www.business.uwa.edu.au/school/disciplines/economics
ECONOMICS DISCUSSION PAPERS - 2015 DP NUMBER
AUTHORS TITLE
15.01 Robertson, P.E. and Robitaille, M.C. THE GRAVITY OF RESOURCES AND THE TYRANNY OF DISTANCE
15.02 Tyers, R. FINANCIAL INTEGRATION AND CHINA’S GLOBAL IMPACT
15.03 Clements, K.W. and Si, J. MORE ON THE PRICE-RESPONSIVENESS OF FOOD CONSUMPTION
15.04 Tang, S.H.K. PARENTS, MIGRANT DOMESTIC WORKERS, AND CHILDREN’S SPEAKING OF A SECOND LANGUAGE: EVIDENCE FROM HONG KONG
15.05 Tyers, R. CHINA AND GLOBAL MACROECONOMIC INTERDEPENDENCE
15.06 Fan, J., Wu, Y., Guo, X., Zhao, D. and Marinova, D.
REGIONAL DISPARITY OF EMBEDDED CARBON FOOTPRINT AND ITS SOURCES IN CHINA: A CONSUMPTION PERSPECTIVE
15.07 Fan, J., Wang, S., Wu, Y., Li, J. and Zhao, D.
BUFFER EFFECT AND PRICE EFFECT OF A PERSONAL CARBON TRADING SCHEME
15.08 Neill, K. WESTERN AUSTRALIA’S DOMESTIC GAS RESERVATION POLICY THE ELEMENTAL ECONOMICS
15.09 Collins, J., Baer, B. and Weber, E.J. THE EVOLUTIONARY FOUNDATIONS OF ECONOMICS
15.10 Siddique, A., Selvanathan, E. A. and Selvanathan, S.
THE IMPACT OF EXTERNAL DEBT ON ECONOMIC GROWTH: EMPIRICAL EVIDENCE FROM HIGHLY INDEBTED POOR COUNTRIES
15.11 Wu, Y. LOCAL GOVERNMENT DEBT AND ECONOMIC GROWTH IN CHINA
15.12 Tyers, R. and Bain, I. THE GLOBAL ECONOMIC IMPLICATIONS OF FREER SKILLED MIGRATION
15.13 Chen, A. and Groenewold, N. AN INCREASE IN THE RETIREMENT AGE IN CHINA: THE REGIONAL ECONOMIC EFFECTS
15.14 Knight, K. PIGOU, A LOYAL MARSHALLIAN?
15.15 Kristoffersen, I. THE AGE-HAPPINESS PUZZLE: THE ROLE OF ECONOMIC CIRCUMSTANCES AND FINANCIAL SATISFACTION
48
15.16 Azwar, P. and Tyers, R. INDONESIAN MACRO POLICY THROUGH TWO CRISES
15.17 Asano, A. and Tyers, R. THIRD ARROW REFORMS AND JAPAN’S ECONOMIC PERFORMANCE
15.18 Arthmar, R. and McLure, M. ON BRITAIN’S RETURN TO THE GOLD STANDARD: WAS THERE A ‘PIGOU-MCKENNA SCHOOL’?
15.19 Fan, J., Li, Y., Wu, Y., Wang, S., and Zhao, D.
ALLOWANCE TRADING AND ENERGY CONSUMPTION UNDER A PERSONAL CARBON TRADING SCHEME: A DYNAMIC PROGRAMMING APPROACH
15.20 Shehabi, M. AN EXTRAORDINARY RECOVERY: KUWAIT FOLLOWING THE GULF WAR
15.21 Siddique, A., Sen, R., and Srivastava, S.
AUSTRALIA-THAILAND TRADE: AN ANALYSIS OF COMPETITIVENESS AND THE EFFECTS OF THE BILATERAL FTA
15.22 Tyers, R. SLOWER GROWTH AND VULNERABILITY TO RECESSION: UPDATING CHINA’S GLOBAL IMPACT
15.23 Arthmar, R. and McLure, M. PIGOU ON WAR FINANCE AND STATE ACTION
15.24 Wu, Y. CHINA'S CAPITAL STOCK SERIES BY REGION AND SECTOR
15.25 Clements, K. and Si, J. ENGEL'S LAW, DIET DIVERSITY AND THE QUALITY OF FOOD CONSUMPTION
15.26 Chen, S. SHIFTS OF DISTORTION AND CORRUPTION OVER LOCAL POLITICAL CYCLES IN CHINA
15.27 Chen, S. THE EFFECT OF A FISCAL SQUEEZE ON TAX NFORCEMENT: EVIDENCE FROM A NATURAL EXPERIMENT IN CHINA
15.28 Jetter, M. BLOWING THINGS UP: THE EFFECT OF MEDIA ATTENTION ON TERRORISM
15.29 Tang, S. MEDIUM-TERM MACROECONOMIC VOLATILITY AND ECONOMIC DEVELOPMENT: A NEW TECHNIQUE
15.30 Alim, A., Hartley, P. and Lan, Y. ASIAN SPOT PRICES FOR LNG OTHER ENERGY COMMODITIES
15.31 Gannon, B., Harris, D., Harris, M., Magnusson, L., Hollingsworth, B., Inder, B., Maitra, P, and Munford, L.
NEW APPROACHES TO ESTIMATING THE CHILD HEALTH-PARENTAL INCOME RELATIONSHIP
15.32 Czaika, M. and Parsons, C. THE GRAVITY OF HIGH SKILLED MIGRATION POLICIES
15.33 Parsons, C., Rojon, S., Samanani, F, and Wettach, L.
CONCEPTUALISING INTERNATIONAL HIGH-SKILLED MIGRATION
15.34 Chen, S. VAT RATE DISPERSION AND TFP LOSS IN CHINA’S MANUFACTURING SECTOR
15.35 Tait, L., Siddique, A. and Chatterjee, I. FOREIGN AID AND ECONOMIC GROWTH IN SUB-SAHARAN AFRICA
49
ECONOMICS DISCUSSION PAPERS - 2016 DP NUMBER
AUTHORS TITLE
16.01 Xu, R., Wu, Y. and Luan, J. ANALYSIS OF FARMERS’ WILLINGNESS TO ADOPT GENETICALLY MODIFIED INSECT-RESISTANT RICE IN CHINA
16.02 Lia, Y., Fan, J., Zhao, D., Wu, Y. and Li, J.
TIERED GASOLINE PRICING: A PERSONAL CARBON TRADING PERSPECTIVE
16.03 Clements, K.W., Lan, Y. and Si, J. UNCERTAINTY IN CURRENCY MISPRICING
16.04 Parsons, C. and Vézina, P.L. MIGRANT NETWORKS AND TRADE:THE VIETNAMESE BOAT PEOPLE AS A NATURAL EXPERIMENT
16.05 Chang, S., Connelly, R. and Ma, P.
WHAT WILL YOU DO IF I SAY ‘I DO’?: THE EFFECT OF THE SEX RATIO ON TIME USE WITHIN TAIWANESE MARRIED COUPLES
16.06 Yu, F. and Wu, Y. BIASES IN PATENT EXAMINATION AND FIRMS’ RESPONSES: EVIDENCE FROM THE PHARMACEUTICAL INDUSTRY
16.07 Fan, J., Li, J., Wu, Y., Wang, S. and Zhao, D.
THE EFFECTS OF ALLOWANCE PRICE ON ENERGY DEMAND UNDER A PERSONAL CARBON TRADING SCHEME
16.08 Golley, J., Tyers, R. and Zhou, Y. CONTRACTIONS IN CHINESE FERTILITY AND SAVINGS: LONG RUN DOMESTIC AND GLOBAL IMPLICATIONS
16.09 McGrath, G. and Neill, K. FOREIGN AND DOMESTIC OWNERSHIP IN WESTERN AUSTRALIA’S GAS MARKET
16.10 Clements, K.W. and Si, J. SIMPLIFYING THE BIG MAC INDEX
16.11 Priyati, R.Y. and Tyers, R. PRICE RELATIONSHIPS IN VEGETABLE OIL AND ENERGY MARKETS
16.12 Wu, J., Wu, Y. and Wang, B. THE GREENNESS OF CHINESE CITIES: CARBON DIOXIDE EMISSION AND ITS DETERMINANTS
16.13 Arslan, C., Dumont, J.C., Kone, Z., Özden, Ç., Parsons, C. and Xenogiani, T.
INTERNATIONAL MIGRATION TO THE OECD IN THE TWENTY-FIRST CENTURY
16.14 Tomioka, K. and Tyers, R. HAS FOREIGN GROWTH CONTRIBUTED TO STAGNATION AND INEQUALITY IN JAPAN?
16.15 Donovan, J. and Hartley, P. RIDING THE IRON ORE CYCLE: ACTIONS OF AUSTRALIA’S MAJOR PRODUCERS
16.16 Czaika, M. and Parsons, C. HIGH-SKILLED MIGRATION IN TIMES OF GLOBAL ECONOMIC CRISIS
16.17 Lefroy, T., Key, J. and Kingwell, R. A LONGITUDINAL EXAMINATION OF BROADACRE FARM SIZE AND PERFORMANCE IN WESTERN AUSTRALIA
16.18 Arthmar, R. and McLure, M. SRAFFA, MYRDAL AND THE 1961 SÖDERSTRÖM GOLD MEDAL
19.19 Azwar, P. and Tyers, R. POST-GFC EXTERNAL SHOCKS AND INDONESIAN ECONOMIC PERFORMANCE
19.20 Chen, A. and Groenewold, N. OUTPUT SHOCKS IN CHINA: DO THE DISTRIBUTIONAL EFFECTS DEPEND ON THE REGIONAL SOURCE?
19.21 Wu, Y., Zhu, X. and Groenewold, N. THE DETERMINANTS AND EFFECTIVENESS OF INDUSTRIAL POLICY IN CHINA: A STUDY BASED ON FIVE-YEAR PLANS